AI engineering case study
Adaptive Medical Sales Coach
Core stack
RAG · MCP · LangGraph · Real-time AI
Delivery approach
Situation
Medical sales teams learned from static decks, product documents, and manager-led role play. Every representative received the same material despite differences in role, territory, product knowledge, and objection handling. Guidance took time to update, and free-form coaching could introduce product claims outside approved regulatory language.
Task
Create coaching that adapted to each representative while grounding product and compliance statements in approved material. The platform needed to support repeated practice at high volume, report skill gaps to managers, accept controlled content updates, and distinguish coaching suggestions from approved claims.
Action
A versioned registry tags each approved source by product, audience, scenario, region, owner, and validity window. The RAG pipeline retrieves only material allowed for the exercise. A LangGraph state machine runs the scenario, records the representative's response, applies a coaching rubric, and chooses the next objection from the gaps it observed. Product facts retain quotes or citations; generated suggestions carry a coaching label and cannot become approved messaging. MCP-style tools let content owners publish, retire, and validate material without an application deployment. Scenarios stay bounded, which makes rubric scores easier to compare and review. Tests cover citation support, prohibited claims, persona fit, agreement with expert reviewers, and regressions between content versions. Monitoring covers session latency, retrieval misses, content age, and repeated skill gaps, with rollback for a faulty content release.
Result
The platform supports more than 50,000 practice sessions per month and contributed to 65% faster onboarding. Exercises change in response to observed gaps, and managers can see recurring skill patterns without reading every conversation. Content owners can update approved material while the versioned tests check that coaching remains within the required boundaries.
Technology stack
- RAG
- MCP
- LangGraph
- Real-time AI