Python Developer for Meeting Data Extraction

We need a Python developer to extract structured data from public meeting archives and align agenda items to video timestamps. The work includes building a reliable pipeline to process meeting records, identify agenda topics, and map them to specific video segments. You should be comfortable handling unstructured data, improving accuracy, and delivering a clean, reusable solution. Experience with similar data extraction projects is important. This is a part-time project with potential for future collaboration. I need a working prototype pipeline built over public government meeting records for two jurisdictions, covering a 24-month historical window. The work: Retrieve meeting records from public portals (Granicus, Legistar, CivicPlus, CivicClerk or PrimeGov). Agendas, item numbers and titles, document packets, minutes, video/audio. The open-source civic-scraper library covers these platforms and is a reasonable starting point. Produce time-coded transcripts with speaker diarisation. Some platforms publish caption files already — use those where available. The core problem: for each agenda item, identify the time range in the meeting recording where that item was discussed. Extract per item: action taken, motion text, vote outcome and individual vote splits. Report measured accuracy of the above across the full corpus — not a sample you select. Deliverables are files and measurements. No UI, no database, no dashboard, no web app. CSV, JSON, Markdown, and source code with a README. I care about measured accuracy and an honest inventory of what failed, far more than about presentation. A pipeline that hits 70% and tells me precisely where the other 30% broke is a success. A polished demo that can't report its own error rate is not. Structured as three fixed-price milestones. Milestone 1 is small and standalone — clean corpus index for one jurisdiction — and either of us can stop there. Full spec provided to shortlisted candidates.