2PointZero Group

FTE Benchmarking Agent - Conversational workforce analysis across the IHC portfolio

Workforce benchmarking across 2PointZero's portfolio companies was a multi-week manual exercise - and the source data was sparse and unevenly disclosed across companies. Stakeholders needed a conversational way to ask FTE questions and trust the answers.

2026

Approach

Built a hybrid-RAG agent that classifies companies into 12 segments, imputes missing workforce fields via iterative LightGBM imputation within each segment, and validates imputed values through a multi-LLM majority-vote consensus. Retrieval is hybrid - semantic cosine over Qwen3 embeddings combined with weighted Postgres tsvector - surfaced via an MCP backend and a Streamlit frontend.

Why segment-aware imputation

A single global imputer would be wrong for both a 50-person agency and a 5,000-person industrial conglomerate. Segmenting first, then imputing within segment, kept the imputed FTE numbers defensible when stakeholders asked the obvious follow-up: "how did you get this number?"

Why a multi-LLM consensus

Because the imputed values feed an executive-facing agent, a single LLM critique was too noisy. Majority-vote consensus across multiple providers gave a meaningful abstention signal: when models disagreed, the agent surfaced uncertainty instead of fabricating confidence.