AI HR Assistant
An AI assistant that answers HR questions from the client's own research library and shows its sources. The client's team tracks usage, content gaps and costs from its admin dashboard.


Platforms
- Web
- iOS
- Android
Overview
Our client, a large US professional association for HR, has a big library of research, articles and courses. We built an AI assistant so its members can simply ask. It answers from that library, cites its sources, and suggests courses, webinars or research worth looking at next. Member-only content stays with members. It runs as a web chat app and as a tab inside the association's member app. We also built the admin dashboard behind it. The team uses it to see what members ask, how they rate the answers, where content is missing, what it costs and whether the content pipeline is healthy.
Project details
- Client
- US enterprise client
- Industry
- Professional association for HR
- Market
- United States
- Engagement
- Feb 2026 to present
- Our role
- AI product partner
- Platform
- Web chat app + a tab in the member app + web admin dashboard
What we built
The assistant
- Streaming answers with tables, charts and formulas
- Inline source citations and a sources panel
- Recommendation cards for courses, webinars, certifications, events and research
- Hybrid search (vector + keyword, RRF fusion) with Cohere reranking
- Tools for catalog search, safe data queries and exact calculations
- Member tiers: free users see free content only, and member-only content stays protected
- Answer survives page reload and continues on another device
- Free-tier quota, spend limits and guardrails (no legal advice, on-topic only)
- Expert Review Mode, where HR experts grade answers to improve quality
- Knowledge base fed by 20+ sources (articles, journals, webinars, events, courses, research reports, scanned PDFs) with automatic incremental updates
AI insights and admin dashboard
- KPIs for users, retention, engagement, satisfaction (CSAT) and thumbs up / down
- Content gaps, the questions members ask that the library doesn't cover yet
- Trending topics and "sources behind bad answers", with flag-to-content-team
- Cost controls with daily spend caps, automatic pause and manual block / release
- AI model picker and release-notes editor (feeds "What's new" in web + app)
- Content pipeline to run, monitor and debug ingestion jobs
Inside the product
















A new chat, with suggested questions and a member's chat history
Technology
Tech stack
- Python
- FastAPI
- pydantic-ai
- PostgreSQL + pgvector
- Azure OpenAI
- Cohere Rerank
- Next.js
- Vercel AI SDK
- TanStack Query
- Recharts
- Tailwind
- Azure



