AI chat interfaces at a federal regulator

TypeScript
Svelte
Node.js
Python
MongoDB/DocDB
LLM APIs

Problem

Economics teams needed to interact with LLM-backed retrieval systems — but generic chat UIs fail on accessibility, lose conversation context, and hide retrieval provenance.

Approach

Designed and built structured chat interfaces supporting RAG-based retrieval and multi-session workflows. Established reliable data flow between frontend and AI APIs for real-time interaction. Partnered with accessibility teams so every interaction pattern met WCAG standards — including screen-reader-usable streaming responses.

Outcome

Reduced retrieval latency and delivered faster, more reliable experiences for research workflows.

Internal work — patterns and architecture discussed; no screenshots.