Practical notes from the engineers who build agents, copilots and AI products.
A single event pipeline with a stable customer identity across storefront, app and messaging is what makes personalisation possible. A customer event pipeline defines the events once, stitches anonymous and known identities, delivers events within seconds, and keeps a replayable history for training and testing.
An eval suite is a scored set of real inputs and expected outputs run on every change; it is the difference between a demo and a product. It replaces opinion with a number, catches regressions before users do, and lets you swap models and prompts safely. Here is how to build one and what it costs.
We ship evaluation suites with every AI feature because a demo proves it can work once and evals prove it keeps working. Every prompt, model and retrieval change we make is run against a graded set of real cases before release, and the client keeps the suite, so the system can be changed safely long after we leave.
An enterprise assistant ingests Drive, Slack, wikis and tickets with permissions, refreshes on change and answers with citations. The build is mostly connectors, access control and evaluation rather than model choice; this guide walks through each layer, the decisions that matter and a realistic team and timeline.
Wrap the legacy system with an API, then add extraction, search, copilots and reporting on the stabilised core. The old system keeps its data and rules; the AI layer reads through the API, writes through the same validated entry points, and every action is policy-gated and logged so nothing bypasses existing controls.
Under DPDP, AI systems need lawful purpose, consent or legitimate use, retention limits, breach notification and a grievance officer. The Act never mentions AI, but any AI system touching personal data of Indian residents is processing under it, and the choices that make compliance easy are cheapest before the build.
Document intelligence parses layout, tables and handwriting from PDFs and photos, extracts fields with confidence scores and routes exceptions. It replaces manual keying for invoices, KYC documents, claims and contracts, and the design decision that matters most is what happens when the model is unsure.
Dispatch batches orders by area and window, proposes assignments with routing rules and lets operations override on a live map. A good engine scores candidate riders every few seconds for each unassigned order, with rules the operations team can read and a shadow mode that proves each change before it goes live.
Good copilot UX streams responses, shows confidence and citations, proposes actions before executing and hands off to the normal UI when unsure. Those four patterns decide whether users trust the copilot after the first week. This article covers each pattern, the entry points that drive use, and pre-launch testing.