AttributionDelivered by our founder before Nexaces was established, as part of a prime contractor's delivery team. Listed as relevant experience, not as a Nexaces contract.
The problem
An AI assistant is only as useful as what it can find. Most of the value is in getting an organization's documents into a form a model can search, and in an interface that lets people come back to what they found.
The work
Built document ingestion from the browser: users upload files in many formats, which are extracted, converted into vector representations and indexed for semantic search. Built modular React components that show model responses, keep conversational context, and manage bookmarks for whole applications and individual knowledge bases.
Connected the React and TypeScript front end to Java Reactor microservices and Llama model inference so conversations update in real time, and kept the design open to new models, vector database providers and ingestion pipelines.
Outcome
Document upload, semantic search and conversation in one React interface, built so new models and data sources can be added without rework.
← All delivery experience