Loading...
Loading...
2026 · Deployed at the client
Guillaume Schonrock · Updated on
An assistant connected to company documents and conversations, with sourced answers and checks before tool execution.
A 2026 client assignment covering discovery, development and deployment of an AI assistant integrated with Microsoft 365. It connects company documents and conversations to sourced answers, with explicit rules for tool execution.
I defined the product vision, roadmap, acceptance criteria and governance rules, then built the Python API and document processing through to deployment in the client environment.
Assistant deployed in a confidential client environment
Deployed at the client
Product and governance
Python / FastAPI
Help staff find information through their existing work channels. Answers need to link back to an identifiable, authorised source.
Scope includes PDF, Word and Excel ingestion, document synchronisation and retrieval across company sources.
The FastAPI and pydantic-ai API provides a reusable entry point for conversation, documents and structured data. Semantic retrieval uses PostgreSQL/pgvector.
Interactions are classified before tool execution. A rejected or invalid decision blocks tool selection. Human review, minimised traces and trace purging are part of the design.
Idempotent processing and loop prevention handle retries and automatic replies.
Deployment in a confidential client environment was confirmed in September 2026. No public access or demonstration is available.
Usage measurements and user feedback are not yet available. Deployment demonstrates technical delivery; adoption and benefit evaluation remain to be carried out.
Let's talk about your product, platform, or operational workflow and see how I can help shape and build it.