P2Predict
Working on P2Predict, a parametric pricing tool for agents, mainly in procurement, but also sales. Point it at your own purchase history and it learns what actually drives the price of a part (supplier, material, spec, region, whatever), then an AI agent can ask it things like "is this $14.20 quote fair" and get back a number with a confidence range and a spec-by-spec breakdown instead of a guess. This can be used to drive a negotiation (Nego targets based on data), or for a value workshop with engineering (reducing this x feature in the spec should save us y% in BOM) or for sales when preparing a bid, etc. I even used it to predict how to create savings in rent as I needed to move apartments in Singapore. This comes out of years driving negotiations at Infineon, Dyson and other companies across the world. It's a CLI under the hood, ships an MCP server so it plugs into Claude, GPT, whatever agent + harness you're already using. Also spent some time lately trying to validate the math outside procurement. Ran it against the public TDC ADMET pharma benchmarks and it placed 7th of 24 on one of the boards, which I wasn't expecting from something built for spend data. 100% local and private, Free for internal company use, source-available: github.com/ahmed-khalil-hafsi/P2Predict