The words we use, explained.
181 terms across AI, product engineering, modernization and the industries we work in. Each one says what it means, who it matters to, why it exists, where it is applied and whether it is a skill you can hire for.
- Personalisation & machine learning
A/B test (controlled test)
An A/B test randomly assigns users to a control group and one or more variants, then compares outcomes, so a change's effect on revenue, conversion or retention is measured rather than assumed.
Read - Industry terms
ABDM (Ayushman Bharat Digital Mission)
ABDM is India's national digital health framework, providing health IDs (ABHA), registries for facilities and professionals, and consent-based health-record exchange that hospitals and health-tech products integrate with.
Read - Industry terms
Adaptive learning
Adaptive learning adjusts what a learner sees next, such as content, practice questions and pacing, based on their demonstrated mastery and behaviour, using models grounded in the institution's own curriculum.
Read - Modernization, ERP/CRM & implementation
Adoption measurement
Adoption measurement tracks whether people are actually using a new system for the work it was meant to replace, through usage data and process indicators, so problems are found and fixed rather than assumed away.
Read - Copilots & AI features
Adoption metrics
Adoption metrics measure whether users actually take up an AI feature and keep using it, tracking activation, weekly active use, acceptance of suggestions, retention over time and the share of work completed through the feature.
Read - Customer service AI
Agent assist
Agent assist is AI that works alongside human support agents rather than replacing them, suggesting answers, surfacing relevant knowledge and customer data, drafting replies and summarising conversations while the human stays in control of what is sent.
Read - AI agents & orchestration
Agent memory
Agent memory is the set of mechanisms by which an AI agent retains information across steps and sessions: the current task's working state, prior interactions with a user, and learned facts, stored outside the model and retrieved when relevant.
Read - AI agents & orchestration
Agentic AI
Agentic AI describes systems designed around AI agents that plan, act and verify across multiple steps and tools to complete goals, as opposed to single-shot models that return one answer to one prompt.
Read - Text-to-SQL & analytics AI
Aggregation pipeline (MongoDB)
A MongoDB aggregation pipeline is a sequence of stages that filters, groups and reshapes documents; for analytics AI it is the target a model must generate instead of SQL when data lives in MongoDB.
Read - AI agents & orchestration
AI agent
An AI agent is software that uses a language model to interpret a goal, decide which tools or systems to call, act on them within set limits, and check the result, rather than only producing text.
Read - Copilots & AI features
AI copilot
An AI copilot is an assistant embedded in a product or workflow that helps a user complete their own task faster, by drafting, summarising, answering and suggesting actions, while the user stays in control of what is done.
Read - Maintenance & AI operations
AI cost per account
AI cost per account is the total inference, retrieval and infrastructure spend attributable to each customer account in a SaaS product over a period, used to check that AI features are priced and used profitably.
Read - Copilots & AI features
AI feature pricing
AI feature pricing is how a SaaS product charges for AI capabilities, choosing between bundling into a higher tier, a per-seat add-on, usage-based credits or outcome-based fees, while covering variable inference cost.
Read - AI strategy & readiness
AI governance
AI governance is the set of policies, controls and records that decide what an AI system may do, who is accountable for it, how it is tested and monitored, and how its decisions can be explained afterwards.
Read - AI strategy & readiness
AI readiness assessment
An AI readiness assessment is a structured review of your data, systems, people and processes that tells you which AI use cases you can actually ship now, and what must change first.
Read - AI strategy & readiness
AI roadmap
An AI roadmap is a sequenced plan of AI projects over two to four quarters, each with a scope, an owner, a measurable outcome and the data or platform work that must land before it.
Read - AI strategy & readiness
AI use-case discovery
AI use-case discovery is the process of finding, describing and ranking the specific tasks in a business where AI could take work off people, scored by value, feasibility and risk rather than novelty.
Read - Voice AI
AI voice agent
An AI voice agent is software that answers or places phone calls, understands natural speech, holds a multi-turn conversation and completes tasks such as booking, verification or payment reminders without a human on the line.
Read - Personalisation & machine learning
Anomaly detection
Anomaly detection identifies events, transactions or metrics that deviate from expected patterns — a suspicious payment, a sensor reading, a sudden drop in orders — so they can be reviewed or blocked before they cause harm.
Read - Modernization, ERP/CRM & implementation
API layer over a monolith
An API layer over a monolith is a service that exposes a legacy system's data and functions through clean, documented endpoints, so new apps and AI agents can use it without touching its internals.
Read - Product engineering & SaaS
API-first design
API-first design is the practice of specifying and building a product's public API before its user interface, so that the web app, mobile apps, partners and AI agents all use the same well-defined contract.
Read - AI agents & orchestration
Approval gate
An approval gate is a workflow step where an AI-proposed action pauses until a named person confirms, edits or rejects it, with the proposal, evidence and reasoning presented so the decision takes seconds.
Read - Product engineering & SaaS
Audit log
An audit log is an append-only record of who did what, to which object, when and from where inside a software product, kept so that security incidents, disputes and compliance reviews can be reconstructed.
Read - AI agents & orchestration
Audit trail
An audit trail for an AI system is the tamper-evident record of every input, retrieved source, model call, tool call, policy decision, human approval and outcome, so any action can be explained and reconstructed afterwards.
Read - Customer service AI
Auto-resolution rate
Auto-resolution rate is the share of customer conversations an AI support agent fully resolves without a human, measured by the customer's problem actually being fixed and not reopened, rather than by the ticket simply being closed.
Read
- Industry terms
Bank statement analysis
Bank statement analysis extracts transactions from PDF or account-aggregator statements, categorises them and derives income, obligations and risk signals that feed credit decisions for lending.
Read - Voice AI
Barge-in
Barge-in is a voice agent's ability to detect that the caller has started speaking while the agent is still talking, stop its own audio immediately and listen, the way a person would in normal conversation.
Read - Copilots & AI features
Beta cohort
A beta cohort is a deliberately chosen group of accounts or users who receive a new AI feature behind a feature flag first, so adoption, quality and cost can be measured and fixed before general release.
Read - AI strategy & readiness
Build vs buy vs integrate
Build vs buy vs integrate is the decision framework for each AI capability: write it yourself, license a product, or connect an existing platform to your systems with a thin custom layer.
Read
- Voice AI
Call recording consent
Call recording consent is the legal and practical requirement to tell callers that a call is being recorded and processed by an AI system, and to obtain or honour their consent, before storing audio or transcripts.
Read - Maintenance & AI operations
Care plan (AMC)
A care plan, or annual maintenance contract, is a fixed monthly agreement covering monitoring, patching, bug fixes, model updates and defined support hours for a live software or AI system after go-live.
Read - Engagement & commercial
Care Plans
Care Plans are our monthly maintenance and support tiers, Essential at $1,000, Standard at $2,500 and Enterprise from $5,250, that keep a delivered system monitored, patched, evaluated and supported after launch.
Read - Modernization, ERP/CRM & implementation
Characterisation tests
Characterisation tests capture what a legacy system actually does today, correct or not, so that any change during modernization can be checked against its existing behaviour.
Read - RAG & knowledge engineering
Chunking
Chunking is splitting source documents into pieces small enough to embed and retrieve precisely but large enough to keep their meaning, with the split strategy chosen by document type.
Read - Personalisation & machine learning
Churn prediction
Churn prediction is a model that estimates, for each customer or account, the probability of leaving within a set window, early enough for the retention team to intervene where it will make a difference.
Read - Copilots & AI features
Citations in UI
Citations in UI are clickable references shown alongside an AI answer that link each claim back to the specific document, section or record it came from, letting users verify the answer in one click.
Read - Industry terms
COD reconciliation
COD reconciliation matches cash collected on delivery against orders, courier remittances and bank credits, identifying shortfalls, delays and disputes automatically so that finance knows exactly what is owed and by whom.
Read - Engagement & commercial
Code and IP ownership
Code and IP ownership at Eazyware means you own everything we build for you: source code, prompts, fine-tuned models, eval suites and infrastructure definitions, transferred to your accounts at handover with no licence-back or lock-in.
Read - Personalisation & machine learning
Cold start
Cold start is the situation where a personalisation system has no behavioural history for a new user or a new item and therefore cannot personalise until it gathers signal or falls back to other information.
Read - Personalisation & machine learning
Collaborative filtering
Collaborative filtering is a recommendation method that predicts what a user will like from the behaviour of users with similar histories, without needing to understand the items themselves.
Read - Industry terms
Collections automation
Collections automation uses voice, WhatsApp and SMS agents to remind borrowers, negotiate within approved rules and record promises to pay, while keeping every contact inside regulatory and consent limits.
Read - Personalisation & machine learning
Computer vision
Computer vision is the branch of machine learning that extracts meaning from images and video — reading documents, inspecting products, counting objects, verifying identity — so software can act on what a camera or scanner captures.
Read - Copilots & AI features
Confidence indicator
A confidence indicator is a UI signal that tells the user how much to trust a specific AI answer, derived from retrieval quality, model agreement or eval-calibrated scores, so uncertain outputs get checked and reliable ones get used.
Read - LLM application engineering
Context window
The context window is the maximum amount of text, measured in tokens, that a language model can consider in a single request, covering the instructions, supplied documents, conversation history and its own answer.
Read - Text-to-SQL & analytics AI
Conversational analytics
Conversational analytics is a multi-turn chat interface over governed business data, where follow-up questions build on earlier answers and results appear inside tools such as Slack, Teams or a SaaS product.
Read - Copilots & AI features
Copilot actions
Copilot actions are the defined operations a copilot can perform in a product on the user's behalf, such as creating a ticket or applying a discount, executed through existing APIs under the user's own permissions and, where risky, behind confirmation.
Read - Maintenance & AI operations
Cost per resolved conversation
Cost per resolved conversation is the total cost of running an AI support agent, including inference, tooling and human escalation, divided by the number of customer conversations it fully resolved without reopening.
Read - Product engineering & SaaS
Crash-free sessions
Crash-free sessions is the percentage of app sessions that end without a crash, tracked per release and platform, and the mobile stability metric most closely tied to ratings, retention and support load.
Read - Customer service AI
CSAT (customer satisfaction score)
CSAT is the customer satisfaction score collected right after a support interaction, usually a 1–5 rating, expressed as the percentage of positive responses; for AI support it is the guardrail showing whether automation helps or merely absorbs customers.
Read - Modernization, ERP/CRM & implementation
Custom CRM
A custom CRM is customer relationship management software built to match one organisation's sales, service or field processes, with its own data model, workflows and integrations, instead of configuring a packaged CRM.
Read - Modernization, ERP/CRM & implementation
Custom ERP
A custom ERP is enterprise resource planning software built around one organisation's actual processes, covering the modules it needs such as inventory, finance, procurement and production, rather than adapting a packaged product.
Read
- Personalisation & machine learning
Data leakage
Data leakage is when information that would not be available at prediction time finds its way into a model's training data, producing impressive test results and a model that fails in production.
Read - Modernization, ERP/CRM & implementation
Data migration and cleansing
Data migration and cleansing moves records from a legacy system into a new platform after deduplicating, correcting and restructuring them, so the new system starts with data people can trust.
Read - Private AI & compliance
Data residency
Data residency is the requirement that certain data be stored and processed within a specific country or region, which determines where AI models, vector stores and logs can physically run.
Read - Engagement & commercial
Dedicated pod
A dedicated pod is a small, stable Eazyware team of engineers assigned to one client on a monthly retainer, working from the client's backlog rather than a fixed project scope.
Read - Personalisation & machine learning
Demand forecasting
Demand forecasting predicts how much of each product or service will be needed at each location and time, so that inventory, staffing and purchasing decisions are made from expected demand rather than last year's guess.
Read - AI strategy & readiness
Discovery sprint
A discovery sprint is a fixed, short engagement, ten working days at Eazyware, that assesses readiness, ranks use cases, tests the riskiest assumption and delivers a scoped, priced plan for the first build.
Read - Product engineering & SaaS
Dispatch engine
A dispatch engine is the software component that assigns incoming jobs, such as deliveries, rides or service calls, to the right worker or vehicle within seconds, using rules and scoring over location, capacity, skills and cost.
Read - RAG & knowledge engineering
Document intelligence (OCR)
Document intelligence is the extraction of structured data from PDFs, scans, photos and forms, combining OCR with layout understanding and language models to produce fields you can validate and act on.
Read - Private AI & compliance
DPDP Act 2023
The Digital Personal Data Protection Act 2023 is India's data-protection law, administered under MeitY, setting consent, purpose-limitation and security obligations for organisations processing personal data, including in AI systems.
Read
- RAG & knowledge engineering
Embedding
An embedding is a list of numbers that represents the meaning of a piece of text, image or record so that similar meanings sit close together and can be compared mathematically.
Read - Customer service AI
Escalation with context
Escalation with context is the handover from an AI support agent to a human that carries the full conversation, the customer's verified identity, what was already tried and a suggested next step, so the customer never has to repeat themselves.
Read - LLM application engineering
Evals (evaluation suite)
An evaluation suite is a repeatable set of test cases, scoring rules and thresholds that measures whether an AI system produces correct, safe output, run on every prompt, model or data change.
Read - Personalisation & machine learning
Event pipeline
An event pipeline is the infrastructure that captures user actions such as views, clicks, purchases and app screens, validates them against a schema and delivers them reliably to the systems that personalise and analyse.
Read
- Product engineering & SaaS
Feature flags
Feature flags are configuration switches that turn functionality on or off for specific users, tenants or percentages at runtime, so code can be deployed before it is released and rolled back without a redeploy.
Read - RAG & knowledge engineering
Fine-tuning
Fine-tuning is further training of an existing language model on your own examples to change its behaviour, style or task performance; it teaches the model how to respond, not what your documents currently say.
Read - Engagement & commercial
Fixed price, fixed date
Fixed price, fixed date is our default commercial model: a locked scope, a quoted price and a delivery date agreed before work starts, with change requests priced separately rather than absorbed.
Read
- Private AI & compliance
GDPR
GDPR is the European Union's data-protection regulation, governing how personal data of people in the EU is collected, processed and transferred, with direct consequences for how AI systems handle that data.
Read - LLM application engineering
Golden dataset
A golden dataset is a curated collection of real, representative inputs paired with verified correct outputs, used as the fixed reference against which an AI system's quality is measured.
Read - Text-to-SQL & analytics AI
Golden question set
A golden question set is a curated list of real business questions with verified correct answers, used to measure and regression-test an analytics AI before and after every change.
Read - Private AI & compliance
GPU inference (vLLM)
GPU inference is running a language model on graphics hardware to answer requests; vLLM is a widely used open-source serving engine that batches requests and manages GPU memory so one card serves many users efficiently.
Read - RAG & knowledge engineering
GraphRAG (knowledge graph RAG)
GraphRAG is a retrieval approach that extracts entities and relationships from documents into a knowledge graph, then answers questions by traversing those connections rather than only matching similar text.
Read - LLM application engineering
Grounding
Grounding is the practice of giving a language model verified source material, such as retrieved documents or database records, and requiring its answers to be based on and traceable to that material.
Read - Modernization, ERP/CRM & implementation
GST e-invoicing integration
GST e-invoicing integration connects business software to India's Invoice Registration Portal so that B2B invoices are reported, assigned an IRN and QR code, and returned to the system automatically.
Read - AI agents & orchestration
Guardrails
Guardrails are the checks around a model's inputs and outputs that block or correct unsafe, off-policy or malformed behaviour, such as prompt injection, data leakage, off-topic responses or invalid tool calls.
Read
- LLM application engineering
Hallucination
A hallucination is output from a language model that is fluent and confident but factually wrong or unsupported by any source, such as an invented policy, figure, citation or product detail.
Read - Customer service AI
Helpdesk integration
Helpdesk integration connects an AI support agent to the ticketing platform the team already uses, such as Zendesk, Freshdesk or Intercom, so the AI reads and writes tickets, escalates into the same queues and appears in the same reports.
Read - Private AI & compliance
HIPAA
HIPAA is the United States law governing protected health information, requiring covered entities and their business associates to safeguard patient data, which shapes how AI can be used with clinical and billing records.
Read - Industry terms
HIS, LIS and PACS
HIS, LIS and PACS are the core hospital systems for patient administration, laboratory results and medical imaging respectively; integrating AI with them means working through HL7, FHIR and DICOM rather than around them.
Read - Modernization, ERP/CRM & implementation
HRMS implementation
HRMS implementation is the delivery of a human resource management system covering employee records, leave, attendance, payroll and performance, including data migration, integrations, policy configuration and phased rollout.
Read - AI agents & orchestration
Human-in-the-loop
Human-in-the-loop is a design in which a person reviews, approves, corrects or takes over specific AI decisions or actions, with the system built so that hand-offs carry full context and human corrections improve it.
Read - RAG & knowledge engineering
Hybrid search
Hybrid search combines keyword (lexical) matching with embedding-based semantic matching in one retrieval step, so results catch both exact terms like part numbers and differently worded questions.
Read - Modernization, ERP/CRM & implementation
Hypercare
Hypercare is the intensive support period immediately after a system goes live, with the delivery team on hand, fast fix cycles and daily issue triage until the platform and its users are stable.
Read
- Copilots & AI features
In-app copilot
An in-app copilot is an AI assistant built directly into a software product's interface, with access to the user's current screen, data and permissions, so it can answer and act in context rather than in a separate chat window.
Read - LLM application engineering
Inference cost
Inference cost is the running expense of getting answers from a deployed model, driven by tokens processed, model tier, request volume and infrastructure, as opposed to the one-off cost of building the system.
Read - Industry terms
Insurance claims triage
Insurance claims triage uses AI to read incoming claims and documents, check completeness and policy coverage, score complexity and fraud risk, and route each claim to fast-track settlement, an adjuster or investigation.
Read - Customer service AI
Intent classification
Intent classification is the step where an AI support system works out what the customer wants (refund, order status, cancel subscription, report a bug) from their free-text or spoken message, so it can route, retrieve and act correctly.
Read - Voice AI
IVR (interactive voice response)
IVR is the automated phone-menu system that greets callers and routes them by keypad presses or short spoken keywords ("press 1 for billing"), the predecessor to conversational AI voice agents.
Read
- RAG & knowledge engineering
Knowledge base
A knowledge base is the curated, versioned collection of documents and answers an AI assistant draws on; its coverage, freshness and ownership decide the quality of every answer more than the model does.
Read - Customer service AI
Knowledge gap report
A knowledge gap report is the periodic output from an AI support agent listing the questions customers asked that it could not answer well because the help centre or knowledge base had no adequate content, ranked by frequency and impact.
Read - Industry terms
KYC document processing
KYC document processing is the automated extraction, validation and cross-checking of identity and address documents such as PAN, Aadhaar and passports, with exceptions routed to humans and every decision logged for audit.
Read
- LLM application engineering
Large language model (LLM)
A large language model is a neural network trained on very large text corpora to predict the next token, which lets it read, summarise, classify, extract and generate text on demand.
Read - Voice AI
Latency budget (turn-taking)
The latency budget is the total time a voice agent may take from the caller finishing a sentence to the agent starting to speak, typically under one second, split across speech recognition, the language model and speech synthesis.
Read - Engagement & commercial
Launch 6
Launch 6 is our six-week AI-Accelerated MVP program, priced from $26,500 to $45,500, that ships a production-ready first version of an AI product or feature to a fixed scope and fixed date.
Read - Industry terms
Learner data protection
Learner data protection covers the consent, minimisation, access and retention rules that apply to student data, especially for minors, under laws such as India's DPDP Act, GDPR and FERPA, and how education products must be built to meet them.
Read - Modernization, ERP/CRM & implementation
Legacy modernization
Legacy modernization is the staged replacement or upgrade of an ageing business system so it can be safely changed, integrated and extended with AI, without stopping the business that runs on it.
Read - Modernization, ERP/CRM & implementation
Lift-and-shift vs re-platform
Lift-and-shift moves an application to new infrastructure unchanged; re-platforming changes parts of it, such as the database or runtime, to use managed services. The choice trades speed against long-term cost and flexibility.
Read - LLM application engineering
LLM observability (tracing)
LLM observability is the practice of recording every model request end to end, including prompt version, inputs, retrieved context, tool calls, output, latency, tokens and cost, so behaviour can be debugged, audited and improved.
Read - Maintenance & AI operations
LLMOps
LLMOps is the set of practices for running language-model applications in production: prompt and model versioning, evaluation on every change, tracing, cost monitoring, drift detection and controlled rollout.
Read
- Personalisation & machine learning
Machine learning model
A machine learning model is a function learned from historical data that maps inputs to predictions — a churn probability, a demand figure, a ranking score — instead of being written as explicit rules.
Read - Product engineering & SaaS
Marketplace liquidity
Marketplace liquidity is the likelihood that a buyer finds what they want and a seller finds a buyer within an acceptable time, measured per category and location rather than across the whole platform.
Read - AI strategy & readiness
Minimum viable product (MVP)
A minimum viable product is the smallest version of a product that real users can use to complete a real job end to end, built to learn what to build next rather than to be complete.
Read - Personalisation & machine learning
MLOps
MLOps is the set of practices and tooling for taking machine learning models from a notebook to production and keeping them working: versioning, deployment, serving, monitoring, retraining and rollback.
Read - AI agents & orchestration
Model Context Protocol (MCP)
Model Context Protocol is an open standard for exposing tools, data sources and prompts to AI applications through a common interface, so an internal system can be connected once and used by any compatible agent or assistant.
Read - LLM application engineering
Model deprecation
Model deprecation is when a provider announces that a specific model version will stop being served on a set date, forcing every application that depends on it to migrate, re-test and possibly re-tune.
Read - Maintenance & AI operations
Model drift
Model drift is the gradual decline in an AI system's accuracy after deployment because the real-world data, user behaviour or underlying model changes while the system's assumptions stay fixed.
Read - Private AI & compliance
Model risk audit
A model risk audit is a structured review of an AI or ML model's purpose, data, performance, limitations and controls, producing evidence that the model is fit for its use and its failure modes are managed.
Read - AI agents & orchestration
Multi-agent system
A multi-agent system splits a complex task among several specialised AI agents, each with its own instructions, tools and limits, coordinated by an orchestrator so that no single agent has to do everything.
Read - LLM application engineering
Multi-model routing
Multi-model routing is an architecture in which each request is sent to the most suitable language model, chosen by task type, difficulty, cost, latency or data-residency rules, rather than one fixed model.
Read - Product engineering & SaaS
Multi-tenant architecture
Multi-tenant architecture runs one deployment of a SaaS product for many customer organisations, with each tenant's data, configuration and limits kept logically separate inside shared infrastructure.
Read
- Text-to-SQL & analytics AI
Narrative insights
Narrative insights are short written explanations generated from query results — what changed, by how much, and what most likely drove it — attached to a chart or report so the reader does not have to interpret it alone.
Read - Text-to-SQL & analytics AI
Natural language reporting
Natural language reporting lets a user describe the report they want in plain words and receive a chart, table or scheduled dashboard, generated from governed data rather than hand-built by an analyst.
Read - Personalisation & machine learning
Next best action
Next best action is a decision approach that chooses, for each customer at each moment, the single most valuable thing to do next — an offer, a message, a nudge or nothing — instead of one campaign for everyone.
Read
- Product engineering & SaaS
Offline-first
Offline-first is a mobile app design where the app reads and writes to a local database as the primary store and synchronises with the server when connectivity allows, so work continues without a network.
Read - Maintenance & AI operations
Onboarding audit
An onboarding audit is the structured review a support partner performs when taking over a system it did not build, covering code, infrastructure, dependencies, security, documentation and risks before committing to support targets.
Read - Private AI & compliance
Open-weight model
An open-weight model is a language or vision model whose trained parameters are published for download, so you can run, fine-tune and deploy it on your own infrastructure under its licence terms.
Read - AI agents & orchestration
Orchestrator
An orchestrator is the component that sequences an agentic workflow: it routes tasks to agents and tools, manages state between steps, enforces limits and retries, and decides when to stop or escalate.
Read
- Industry terms
Peak-season readiness
Peak-season readiness is the engineering and operational preparation that lets a retail or consumer platform handle festive-sale traffic and order volumes many times its normal load without outages, slow checkouts or broken fulfilment.
Read - Voice AI
Per-minute usage pricing
Per-minute usage pricing is the model where an AI voice agent's running cost is charged by the minute of call time, bundling speech recognition, language model, speech synthesis and telephony costs into one rate, typically $0.05 to $0.15 per minute.
Read - RAG & knowledge engineering
Permission-aware retrieval
Permission-aware retrieval ensures an AI assistant only retrieves and answers from documents the requesting user is entitled to see, enforcing source-system access rights at query time rather than in the prompt.
Read - RAG & knowledge engineering
pgvector
pgvector is an open-source PostgreSQL extension that stores embeddings as a native column type and supports indexed similarity search, letting teams run vector retrieval inside the database they already operate.
Read - Modernization, ERP/CRM & implementation
Phased go-live
A phased go-live brings a new platform into production one module, site or user group at a time, each with its own migration, training and hypercare, instead of switching everything on in a single event.
Read - Private AI & compliance
PII redaction
PII redaction is the automated detection and removal or masking of personally identifiable information, such as names, phone numbers, Aadhaar or account numbers, from text before it is stored, logged or sent to a model.
Read - AI agents & orchestration
Planner–worker–reviewer pattern
The planner–worker–reviewer pattern is a multi-agent design where one agent breaks a goal into steps, worker agents execute them with narrow tools, and a separate reviewer checks results against the goal and rules before commit.
Read - AI agents & orchestration
Policy-gated action
A policy-gated action is an agent action that runs only after deterministic code checks it against written business rules, such as refund limits or eligibility conditions, with anything outside the rules blocked or routed to a person.
Read - LLM application engineering
Prompt engineering
Prompt engineering is the practice of writing and structuring the instructions, examples and context given to a language model so that it produces consistent, correct output for a defined task.
Read - LLM application engineering
Prompt versioning
Prompt versioning is the practice of storing prompts in source control with a version identifier, a changelog and evaluation results, so every change is reviewable, testable and reversible like code.
Read - AI strategy & readiness
Proof of concept (POC)
A proof of concept is a short, bounded build that tests whether an AI approach works on your real data against a pre-agreed pass mark, producing evidence rather than a demo.
Read - Engagement & commercial
ProofRun
ProofRun is our three-week AI POC Sprint, priced from $6,250 to $10,500, that tests one use case against your real data with a written eval suite, so you know whether it works before funding a build.
Read - Product engineering & SaaS
Push notifications
Push notifications are messages a server sends to a user's device through Apple's APNs or Google's FCM, appearing outside the app, used for time-sensitive updates and re-engagement when permission is granted.
Read
- Product engineering & SaaS
RBAC (role-based access control)
RBAC is an authorisation model where permissions are attached to roles such as admin, manager or viewer, and users are granted roles, so what someone can do depends on their role rather than individual settings.
Read - Industry terms
RBI outsourcing guidelines
RBI outsourcing guidelines set the conditions under which regulated Indian financial institutions can use third parties, including technology vendors, covering due diligence, data protection, audit access, localisation and the institution's continued accountability.
Read - Product engineering & SaaS
React Native vs native
React Native vs native is the choice between building one JavaScript codebase that renders native components on iOS and Android, or building two separate apps in Swift and Kotlin with full platform access.
Read - Personalisation & machine learning
Real-time ranking
Real-time ranking re-orders recommendations, search results or content within a single session, using the user's actions from seconds ago rather than a profile computed overnight.
Read - RAG & knowledge engineering
Recall, precision and groundedness
Recall, precision and groundedness are the three core measures of a RAG system: whether retrieval found the right passages, whether it avoided wrong ones, and whether the answer is supported by what was retrieved.
Read - Personalisation & machine learning
Recommendation engine
A recommendation engine is a system that ranks products, content or actions for each user based on their behaviour and that of similar users, so what they see first is what they are most likely to want.
Read - Engagement & commercial
ReCore
ReCore is our eight-to-sixteen-week Legacy-to-AI Modernization program, priced from $31,500 to $105,000 and above, that modernises a legacy system incrementally and adds AI capability without stopping the business.
Read - Customer service AI
Reopen rate
Reopen rate is the percentage of support conversations marked resolved that the customer comes back on within a set window, usually seven days; for AI agents it is the check that a "resolved" conversation actually fixed the problem.
Read - RAG & knowledge engineering
Reranking
Reranking is a second pass that re-scores the top candidates from initial retrieval with a more accurate model, so the few passages actually handed to the language model are the best ones.
Read - RAG & knowledge engineering
Retrieval-augmented generation (RAG)
RAG is a pattern where a system fetches relevant passages from your own documents or data first, then has a language model write an answer grounded in those passages rather than in its training memory.
Read - Text-to-SQL & analytics AI
Row-level security (RLS)
Row-level security restricts which rows of a table a user can see based on who they are, so an analytics AI answering their question can only ever draw on data they are entitled to.
Read - AI agents & orchestration
RPA vs AI agent
RPA replays fixed, scripted steps on user interfaces or APIs and breaks when inputs vary; an AI agent interprets the goal and the input, chooses actions and handles variation, with policy gates and evals controlling the added flexibility.
Read
- AI strategy & readiness
Scope lock
Scope lock is the point in a fixed-price, fixed-date project where the deliverables are written down and agreed, after which changes go through an explicit swap or a separate phase rather than silently expanding the build.
Read - Maintenance & AI operations
Security patching cadence
Security patching cadence is the agreed schedule and urgency rules for applying security updates to a production system's dependencies, operating systems, containers and models, including emergency handling for critical vulnerabilities.
Read - Private AI & compliance
Self-hosted LLM
A self-hosted LLM is a language model you run on infrastructure you control, such as your own GPUs or your cloud account, so prompts and outputs never pass through a third-party API.
Read - LLM application engineering
Semantic caching
Semantic caching stores model responses and serves them again for new requests whose meaning matches a previous one closely enough, judged by embedding similarity, rather than requiring an exact text match.
Read - Text-to-SQL & analytics AI
Semantic layer
A semantic layer is a curated description of your data — business names, metric definitions, joins and allowed filters — that sits between raw tables and whoever, human or model, is asking questions.
Read - RAG & knowledge engineering
Semantic search
Semantic search retrieves results by the meaning of a query rather than its exact words, using embeddings so that differently phrased questions and documents match when they are about the same thing.
Read - AI agents & orchestration
Shadow mode
Shadow mode runs an AI agent on live traffic without letting it act: it produces decisions alongside the human process, and the two are compared to measure agreement before any autonomy is granted.
Read - Engagement & commercial
Shadow-to-autonomy rollout
Shadow-to-autonomy rollout is our staged way of launching AI agents: the agent first runs silently alongside humans, then acts with approval, then acts alone within policy limits, advancing only on measured performance.
Read - Product engineering & SaaS
Shell-and-modules architecture
Shell-and-modules architecture splits a large app into a host shell that owns identity, payments, navigation and shared services, and independent modules that each deliver one service and can be built and released separately.
Read - Maintenance & AI operations
SLA: response vs resolution
In a support SLA, response time is how quickly a reported issue is acknowledged by a qualified person, while resolution time is how quickly it is fixed or worked around; the two are distinct commitments.
Read - Private AI & compliance
Sovereign AI
Sovereign AI describes AI capability, models, infrastructure and data, that remains under the legal and operational control of a nation or organisation rather than depending on foreign providers.
Read - Voice AI
Speech-to-text (ASR)
Speech-to-text, or automatic speech recognition (ASR), converts spoken audio into a text transcript in real time so a language model can understand a caller; its accuracy on accents, noise and Indian languages sets the ceiling for any voice agent.
Read - Engagement & commercial
Sprint Zero
Sprint Zero is our ten-working-day AI Discovery Sprint, priced at $3,250 or ₹2,00,000, that turns a vague AI ambition into a ranked use-case list, a scoped build plan and a straight answer on whether to proceed.
Read - Product engineering & SaaS
SSO (SAML, OIDC)
SSO lets a customer's staff sign in to your SaaS product with their company identity provider, using the SAML or OIDC protocols, so access is managed centrally and revoked when someone leaves.
Read - Modernization, ERP/CRM & implementation
Strangler pattern
The strangler pattern replaces a legacy system gradually by routing individual functions to new services behind a facade, until the old system handles nothing and can be switched off.
Read - LLM application engineering
Streaming
Streaming is delivering a model's response to the user token by token as it is generated, rather than waiting for the full answer, so the interface feels responsive even when generation takes several seconds.
Read - LLM application engineering
Structured output
Structured output is a technique where a language model is constrained to return data in a defined schema, such as JSON with fixed fields and types, so software can consume the result without parsing free text.
Read - Product engineering & SaaS
Subscription billing
Subscription billing is the system that charges customers on a recurring schedule for plans, seats and usage, handling trials, upgrades, proration, taxes, failed payments and invoices in each currency and jurisdiction.
Read - Product engineering & SaaS
Super app
A super app is a single mobile application that bundles many services, such as payments, commerce, bookings and messaging, around one identity, one wallet and one home screen, often with third-party mini-apps.
Read
- Industry terms
Telehealth platform
A telehealth platform lets patients consult clinicians remotely through video, voice or chat, with scheduling, e-prescriptions, payments, records and consent handled in a system built for clinical and regulatory requirements.
Read - Voice AI
Telephony integration (SIP, Twilio, Exotel)
Telephony integration connects an AI voice agent to real phone calls: via a SIP trunk into your existing PBX or contact centre, or via a cloud provider such as Twilio or Exotel supplying numbers, call control and audio streaming.
Read - Product engineering & SaaS
Tenant isolation
Tenant isolation is the set of controls in a multi-tenant product that guarantee one customer's data, performance and configuration cannot be seen, changed or degraded by another customer.
Read - Voice AI
Text-to-speech (TTS)
Text-to-speech (TTS) turns the language model's written reply into natural-sounding spoken audio, streamed to the caller sentence by sentence so the voice agent can start speaking before the whole answer is generated.
Read - Text-to-SQL & analytics AI
Text-to-SQL
Text-to-SQL is the technique of having a language model translate a plain-English question into a database query, run it against your data and return the result as a table or answer.
Read - Customer service AI
Ticket deflection
Ticket deflection is the percentage of support enquiries that never become a ticket for a human agent because a help centre, chatbot or AI agent handled them first; it measures avoided workload, not whether the customer was actually helped.
Read - Engagement & commercial
Time and materials
Time and materials is an engagement model where you pay for the hours and resources actually used rather than a fixed outcome, which suits open-ended work but shifts delivery risk onto the buyer.
Read - LLM application engineering
Token
A token is the unit of text a language model reads and writes, typically a word fragment of a few characters, and the unit in which context limits and usage-based pricing are measured.
Read - AI agents & orchestration
Tool calling (function calling)
Tool calling is the mechanism by which a language model requests that the application run a named function with structured arguments, such as look up an order or issue a refund, and receives the result to continue reasoning.
Read - AI strategy & readiness
Total cost of ownership (TCO)
Total cost of ownership is the full cost of an AI system over its life: build or licence, inference and infrastructure, integration, evaluation, monitoring, retraining, support and the people who run it, not just the initial price.
Read
- Personalisation & machine learning
Uplift
Uplift is the measured incremental change in an outcome — revenue, conversion, retention — caused by a personalisation or intervention, relative to what would have happened without it.
Read - Product engineering & SaaS
Usage metering
Usage metering is the accurate counting of what each customer consumes in a software product, such as API calls, seats, messages or AI tokens, so that it can be billed, limited and reported.
Read
- RAG & knowledge engineering
Vector database
A vector database stores embeddings alongside metadata and answers nearest-neighbour queries quickly, so a system can find the passages, products or records most similar in meaning to a query.
Read - Voice AI
Voice biometrics
Voice biometrics authenticates a caller by comparing the characteristics of their voice against a stored voiceprint, allowing identity verification during a call without PINs or security questions; it is sensitive biometric data with strict consent rules.
Read - Private AI & compliance
VPC deployment
A VPC deployment runs an AI system inside your own virtual private cloud, an isolated network within your AWS, Azure or GCP account, so models, data stores and logs stay under your account's controls.
Read
- Product engineering & SaaS
Webhooks
Webhooks are HTTP requests a system sends to a subscriber's URL when an event happens, such as a payment succeeding or an order shipping, so other systems react in real time without polling.
Read - Industry terms
WhatsApp commerce
WhatsApp commerce is selling and serving customers inside WhatsApp: catalogue browsing, orders, payments, returns and reorders handled through the Business API, often by an AI agent connected to the brand's order and inventory systems.
Read - AI agents & orchestration
Workflow engine
A workflow engine is software that runs multi-step processes reliably over time, persisting state, retrying failed steps, handling timeouts and waiting for human or external input, so agent workflows survive crashes and delays.
Read
- Private AI & compliance
Zero data egress
Zero data egress is an architecture guarantee that customer or business data processed by an AI system never leaves your controlled network boundary, including to model providers, telemetry services or vendors.
Read - Modernization, ERP/CRM & implementation
Zero-downtime cutover
A zero-downtime cutover switches production traffic from an old system to a new one without a maintenance window, using parallel running, incremental data sync and a rehearsed, reversible switch.
Read
Browse by category
AI readiness assessment · AI use-case discovery · AI roadmap · Build vs buy vs integrate · Total cost of ownership (TCO) · Proof of concept (POC) · Minimum viable product (MVP) · Scope lock · Discovery sprint · AI governance
AI Strategy & DiscoveryAI agent · Agentic AI · Multi-agent system · Planner–worker–reviewer pattern · Orchestrator · Tool calling (function calling) · Model Context Protocol (MCP) · Policy-gated action · Human-in-the-loop · Shadow mode · Guardrails · Approval gate · Audit trail · Agent memory · Workflow engine · RPA vs AI agent
Multi-Agent Systems & Workflow OrchestrationAI voice agent · IVR (interactive voice response) · Speech-to-text (ASR) · Text-to-speech (TTS) · Latency budget (turn-taking) · Barge-in · Telephony integration (SIP, Twilio, Exotel) · Call recording consent · Per-minute usage pricing · Voice biometrics
AI Voice Agents (multilingual)Auto-resolution rate · Ticket deflection · Escalation with context · Agent assist · Knowledge gap report · Intent classification · CSAT (customer satisfaction score) · Reopen rate · Helpdesk integration
AI Customer Service AgentsLarge language model (LLM) · Prompt engineering · Prompt versioning · Evals (evaluation suite) · Golden dataset · Hallucination · Grounding · Structured output · Context window · Token · Inference cost · Multi-model routing · Streaming · LLM observability (tracing) · Semantic caching · Model deprecation
LLM Application DevelopmentRetrieval-augmented generation (RAG) · Embedding · Vector database · Hybrid search · Chunking · Reranking · GraphRAG (knowledge graph RAG) · Document intelligence (OCR) · Permission-aware retrieval · Recall, precision and groundedness · pgvector · Semantic search · Fine-tuning · Knowledge base
Retrieval & Knowledge EngineeringAI copilot · In-app copilot · Copilot actions · Confidence indicator · Citations in UI · Beta cohort · AI feature pricing · Adoption metrics
AI Copilot Development for SaaSText-to-SQL · Semantic layer · Golden question set · Row-level security (RLS) · Natural language reporting · Aggregation pipeline (MongoDB) · Conversational analytics · Narrative insights
Natural Language Data QueryingRecommendation engine · Collaborative filtering · Cold start · Real-time ranking · Next best action · Event pipeline · A/B test (controlled test) · Uplift · Machine learning model · Demand forecasting · Churn prediction · Anomaly detection · Data leakage · MLOps · Computer vision
Personalization EnginesSelf-hosted LLM · Open-weight model · GPU inference (vLLM) · Zero data egress · VPC deployment · DPDP Act 2023 · GDPR · HIPAA · Data residency · Sovereign AI · Model risk audit · PII redaction
Agentic AI Solutions (self-hosted)Multi-tenant architecture · Tenant isolation · SSO (SAML, OIDC) · RBAC (role-based access control) · Audit log · Usage metering · Subscription billing · API-first design · Webhooks · Feature flags · Super app · Shell-and-modules architecture · Marketplace liquidity · Dispatch engine · Offline-first · React Native vs native · Push notifications · Crash-free sessions
SaaS / Cloud-native Application DevelopmentLegacy modernization · Strangler pattern · Characterisation tests · API layer over a monolith · Zero-downtime cutover · Lift-and-shift vs re-platform · Custom ERP · Custom CRM · GST e-invoicing integration · Phased go-live · Data migration and cleansing · Hypercare · Adoption measurement · HRMS implementation
Legacy-to-AI Modernization ProgramCare plan (AMC) · SLA: response vs resolution · LLMOps · Model drift · Security patching cadence · Onboarding audit · Cost per resolved conversation · AI cost per account
Software Maintenance & SupportKYC document processing · Bank statement analysis · Collections automation · RBI outsourcing guidelines · Insurance claims triage · HIS, LIS and PACS · ABDM (Ayushman Bharat Digital Mission) · Telehealth platform · WhatsApp commerce · Peak-season readiness · Adaptive learning · Learner data protection · COD reconciliation
IndustriesFixed price, fixed date · Time and materials · Dedicated pod · Sprint Zero · ProofRun · Launch 6 · ReCore · Care Plans · Code and IP ownership · Shadow-to-autonomy rollout
Fixed-price programs