Samielakkad/AI-Product-Management-Frameworks

AI + Product Management · reusable PM frameworks for shipping LLM products in low-resource markets # AI + Product Management · pm-frameworks-darija > Reusable PM frameworks extracted from shipping a Darija-language conversational marketplace (jak.ma) and from the Baidu ERNIE Mentor Program (October–December 2025). Pricing taxonomy, evaluation rubric, rater calibration, verifier philosophy, multimodal decision matrix. --- ## Why this exists Most PM frameworks you find online are written for B2B SaaS in English-speaking markets with deep incumbents and clean data. Almost none survive contact with a low-resource dialect, a market without published price discovery, and an AI stack that will fabricate confidently in a language you cannot easily evaluate. This repository is the set of frameworks I had to invent to ship one specific product. I publish them so the next person does not have to invent them again. The frameworks here have been load-bearing on: - **jak.ma** — production Moroccan service marketplace, 3,400 daily queries, 0.7% verifier rejection rate - **Baidu ERNIE Mentor Program** — three months of evaluation work on a frontier Chinese LLM (October–December 2025) If you are building in a low-resource language, an unstandardized market, or a domain where LLMs will fabricate plausibly, the frameworks should transfer with minimal adaptation. --- ## What's inside | File | What it is | When to use it | |---|---|---| | `pricing-taxonomy.md` | How to build a verifiable price table from scratch in a market with no published prices | Any marketplace where the model will be asked "how much does X cost?" | | `evaluation-rubric.md` | Five-dimension rubric (factuality, naturalness, trade-fit, price-fairness, geographic) with 0–4 anchors | Any conversational system where "correct" is multi-dimensional | | `calibration-protocol.md` | How to calibrate human raters to Krippendorff α ≥ 0.7 in two sessions | Before you trust any rater scores | | `verifier-philosophy.md` | When to use a deterministic verifier vs LLM-as-judge | Designing the contract between your mode …