.png&w=256&q=75)
🤓 Latest Submissions
HR AI Command Centre Project (Govern.ai)
We built Govern.ai as a governed HR operations platform with a clear separation between the control plane and AI inference. The Next.js dashboard is the operator experience for chat, cases, approvals, audit, screening, attrition, and analytics. It talks to a FastAPI backend that orchestrates specialized HR workflows. The backend is deterministic-first: routine triage, approvals, and safety checks continue even if an LLM is unavailable. For policy questions, we use RAG over Postgres with pgvector, so answers are grounded in retrieved policy content and include citations. Sensitive workflows stop at human approval checkpoints, while every action and decision is recorded in a redacted, append-only audit trail. For intelligence, the same provider-neutral model factory can route live requests to Fireworks AI for hosted inference, or to an optional AMD ROCm/vLLM deployment for self-hosted acceleration. Model IDs, endpoints, and credentials are injected at runtime and allowlisted, not hardcoded. That gives us useful AI when configured, but keeps HR decisions explainable, reversible, and under human control.
13 Jul 2026
