Converts reviews, one-on-ones, and recognition into a queryable evidence layer.
Director Briefing
Making talent signals queryable without turning people into a hidden scoring model.
A retrieval-oriented knowledge layer for talent conversations, reviews, and evidence-backed management questions.
Uses both exact-text and meaning-based retrieval instead of relying on one search mode.
Exposes bounded workflow tools so LLMs can summarize before drilling into raw records.
Warms retrieval at startup and keeps diagnostics protected so LLM access stays reliable without widening data exposure.
Keeps the public briefing surface separate from the protected data and query plane.
What this proves
A practical way to query talent evidence without flattening it into a score.
Talent KB is a focused retrieval layer over existing systems. It proves that reviews, one-on-ones, and recognition can be normalized, searched, and consumed by LLMs in a way that is faster than the native workflow and more disciplined than freeform prompting against raw exports.
Problem
The source data is valuable, but the native workflow is slow and fragmented.
Paycor holds reviews, one-on-one notes, and recognitions, but the default experience is still page-by-page. That is workable for one record at a time and weak for cross-person or cross-period questions.
What We Built
Talent KB is a retrieval layer, not a replacement HR system.
It captures authorized data, normalizes it into a structured knowledge base, and exposes protected workflows through REST and MCP so LLMs can answer plain-English questions against evidence.
Design Decision
The operating model is evidence first, not hidden scoring.
The system is built around retrieval, digests, and drill-down. It does not assign a secret score to employees or attempt to automate stack ranking.
Current Scope
The proof of concept is intentionally bounded.
The current implementation validates ingestion, normalization, retrieval, access control, and LLM consumption against a bounded reporting chain before any broader rollout.
System flow
The proof of concept in one view
Architecture
Stack, diagrams, MCP, and integration model
How Paycor ingestion, retrieval, APIs, and MCP fit together.
Use Cases
What managers, directors, and HR can ask
Role-based workflows and plain-English prompts that map to the current system.
Limitations + Next
Where the edges are, and what expansion could unlock
The realistic constraints, risks, and high-leverage next steps.