Enterprise healthcare analytics platform

Analysts needed live, filterable views of healthcare-professional (HCP) data across specialties and countries, and the prep work behind every cycle was manual.

1. 01Shipped 7 production modules, including an AI-powered Insights Agent, a live dashboard with dynamic country and specialty filtering, an agent-execution observability layer, Grant Review, Scientific Experts, and Impact & Outcomes.
2. 02Built the real-time backend (FastAPI and PostgreSQL in Docker) that queries live HCP databases across 28 specialties, with sentiment scoring and omnichannel HCP profiling.
3. 03Moved the model layer from local Ollama to Azure OpenAI, and picked the model with a 30-case, 6-category, 4-model benchmark harness instead of guessing.
4. 04Automated 3 manual ETL processes with LangChain and CrewAI agents.
5. 05Instrumented agent observability: execution tracing, failure monitoring and token-level cost tracking. Cut LLM inference cost through prompt and context optimization.
6. 06Fixed a frontend race condition with request cancellation (AbortController) rather than debouncing, and traced a production outage to table setup that only ran as a script, replacing it with idempotent seeding at app startup.

• 7 modules in production.
• Live data across 28 specialties.
• Analyst prep cut from about 4 hours to under 20 minutes per cycle.
• Lower LLM inference cost from prompt and context optimization.

• →Turn the observability layer into automated evals on agent output quality.

Client and data are described generically.