US federal contractor · GovCon · AI
Ask the business anything, in plain English
A natural-language assistant that lets staff question contracts, customers and tasks without writing SQL or waiting on a report.
- Services
- AI & Automation
- Stack
- OpenAI AWS Bedrock LangChain Embeddings n8n Node.js PostgreSQL Angular
(01) — The challenge
Answers about contracts, customers and workload sat in the database, but only people who could write queries could reach them. Every question became a request to the tech team.
(02) — What we did
- 01Designed a six-step multi-agent pipeline: router, domain, schema, SQL generator, SQL validator and summariser
- 02Grounded the model in the live schema so it only asks about tables and fields that exist
- 03Added a validator agent that checks every generated query before it runs
- 04Returned plain-English summaries alongside the underlying data
(03) — What shipped
The work behind the outcome.
Users ask operational questions in natural language and receive a business-readable answer with the source data close at hand.
The assistant narrows each request to the right domain before generating SQL, reducing irrelevant joins and keeping answers aligned to the live application schema.
Validation happens before execution, giving the team a controlled path to AI-assisted reporting without opening unrestricted database access.
(04) — Results