Selected work · BrisAI studio system
Business knowledge, turned into a useful conversation.
Ask BrisAI turns an approved knowledge collection into a conversational service. Visitors can explore AI opportunities, follow sources and move from a question to a project conversation. BrisAI engineered the application across its interface, retrieval service, structured answers and operational controls.
The business problem
A business knowledge collection is only useful when people can find the part relevant to their question. A conventional FAQ cannot anticipate every combination of context and follow-up.
BrisAI needed an application that could explain its work, answer practical questions and connect an interested visitor with a suitable next step.
Inside the experience
How the system works
01
Ask in plain English
A focused web interface accepts a question and preserves the context of a follow-up.
02
Retrieve the relevant knowledge
Vector and full-text search locate approved source material for the response.
03
Answer, cite and guide
The application validates its references and offers relevant information or a route to a project conversation.
The engineering behind it
A custom FastAPI service retrieves from a curated corpus using vector and full-text search. It combines those results before generating a structured answer.
The response schema separates answer text, citations, opportunity references and handoff. The service resolves references against retrieved material before the website streams the reply.
The surrounding software manages follow-up context, personal-detail redaction, durable rate and budget limits, feedback and operational tracing.
- Application
- Python / FastAPI service and TypeScript / Next.js interface
- Knowledge
- Curated corpus, hybrid retrieval and citation resolution
- Operations
- Postgres limits, budget controls and tracing
Evidence and operation
A recorded evaluation, with the context intact.
89 authored cases across six suites, recorded on 21 September 2026. Expected-source retrieval checks apply to a 54-case subset.
- Cases without invented citation identifiers
- 89 / 89
Every citation identifier in these recorded responses resolved against the allowed material. This checks citation validity, not complete answer correctness.
Run 2026-09-21 · production-gpt5-neon · citation.none_invented
Dated in-house test of one configuration; see the complete report and its three failures.
- Expected source found in top eight retrieval results
- 54 / 54
Cases with a declared expected source in the retrieval subset. The remaining cases do not belong to this denominator.
Run 2026-09-21 · production-gpt5-neon · retrieval any-hit@8
- The report records three failures: one intent classification, one handoff and one overclaim check.
- Model-judge fully-grounded scoring was 74.2%; successful citation validation is not a guarantee that every answer is fully grounded.
- The dataset was authored for this application. These results do not establish customer outcomes or performance after subsequent changes.
Approved information
Answers depend on the curated corpus and its freshness. New business copy must be ingested when it changes.
A route to a person
A knowledge assistant supports discovery; project commitments and individual advice remain a conversation with BrisAI.
Where this could take your business
Customer knowledge services, internal search and AI features that need a useful interface around controlled information.
Practical guides
- What makes a useful AI application? Inside Ask BrisAI
How BrisAI combines a useful interface, retrieval, structured answers and operational controls to build a custom AI application.
Your next move
Let’s build what moves your business forward.
Bring a problem, an idea or an ambition. Start with a free 20-minute conversation to explore the possibilities and a useful next step.