Practical notes from the engineers who build agents, copilots and AI products.
Custom operational software handles bed management, consent, audits and reporting, integrated with HIS and billing. It replaces the spreadsheets and WhatsApp groups that run most wards, gives every bed and consent form a live state, and produces audit evidence on demand instead of in a scramble before each inspection.
Choose per task by benchmarking on your data for accuracy, latency and cost; most production systems route between two or three providers. No single model wins every task, leaderboards do not predict your workload, and a routing layer keeps the choice reversible. Here is the method we use and what it costs.
Benchmark on your tasks; open-weight models often match on retrieval and extraction and lag on hard reasoning, which routing can cover. Public leaderboards tell you which model is best at leaderboards. A golden set of a few hundred of your own tasks, scored the same way for every candidate, tells you what to ship.
On-device models handle voice, vision and quick tasks offline; cloud models handle reasoning; route by latency and privacy. In practice: small models on the phone for wake words, transcription, OCR and classification, a cloud model for judgement, and a router deciding per request by network, battery and sensitivity.
Offline-first means a local database, queued actions replayed in order, and written conflict rules; it decides the data model, so start there. The app reads and writes locally, a sync engine pushes and pulls in the background, and every record type has a rule for when two edits collide. Retrofitting this is a rewrite.
Next-best-action models decide the one intervention most likely to help each customer now: an offer, a nudge or a save. Next best action AI ranks every permitted action against a business objective, respects eligibility and contact rules, and is only as good as the feedback loop recording what happened next.
Natural-language reporting maps a question onto a semantic layer, runs a validated read-only query and returns a chart with definitions. Without the semantic layer it is text-to-SQL guessing at what "active customer" means. This article covers the architecture, the guardrails, accuracy and the build.
Multilingual support agents detect language, answer in it and hand off with a translated summary, covering Hindi and regional languages alongside English. The hard parts are code-mixed messages, script variation, per-language evaluation and a help centre that only exists in English. Here is what to plan for.
Multilingual AI needs per-language benchmarks and code-switch handling; the same discipline that works for Indian languages works globally. Build evaluation sets from real conversations in each language, benchmark models per language, treat mixed-language turns as normal, and let native speakers judge before launch.