Demonstration project
Kiln
An applied intelligence product that helps engineers read decades of technical documentation with confidence.
Overview
Kiln operates and maintains energy infrastructure. Its engineers work from a technical archive that is extensive, valuable, and difficult to search.
Challenge
A wrong answer is worse than no answer. The product needed to be genuinely useful without ever presenting an unsupported claim as fact.
Approach
We began with evaluation rather than modelling: defining what a good answer looks like, and how a bad one would be detected, before writing product surface.
Engineering
Retrieval is grounded strictly in the document set, every response carries its sources, and confidence is expressed in the interface rather than hidden in a score.
Outcome
Engineers now reach a cited answer in a fraction of the time previously spent searching, and the archive has become a working asset rather than a reference library.
Kiln is a demonstration project; no figures are published.
Technology
- Python
- TypeScript
- Next.js
- PostgreSQL
- pgvector
- AWS Bedrock
