Voice assistant for P&L monitoring and contract ingestion in trading
P&L queried by voice and PDF contracts ingested automatically, figures read from the database.
The challenge
Querying a position or a P&L required going through complex screens or the middle-office team, and integrating contracts received as PDFs remained entirely manual.
Our response
We delivered a two-part proof of concept. First part: voice querying of the P&L, where the spoken question is transcribed, translated into a structured query over the positions, then returned with the exact figures from the database, with no free generation of values. Second part: automatic ingestion of PDF contracts, with extraction of key terms into a structured schema and human validation before storage. The whole thing was built to be demonstrable in front of a business committee.
Key points
Spoken question translated into a structured query, figures read from the database and never generated
Extraction of contractual terms into a constrained business schema
Human validation loop before writing, suited to a regulated environment
From business scoping to client demonstration in a few weeks
Technical stack
- Voice transcription (Whisper large-v3)
- Function calling / text-to-query over a positions database
- RAG over PDF contracts with structured extraction
- pgvector vector store
- Self-hosted open-source LLM
- Answer traceability and query logging
Sector
Trading & commoditiesLet's talk about your project
Tell us about your business challenge and we'll explore together how a tailored AI solution could support it.