Phase 5d: Council Member Profiling

[kpeatt]

Overview

Generate rich, AI-synthesized council member profiles from transcripts, votes, and motions — surfacing policy positions, legislative effectiveness, speaking patterns, voting alignment by topic, and notable moments.

Tasks

• Schema: Create person_profiles table, agenda_item_topics table. Add pronouns and short_bio to people.
• Topic taxonomy: Define MUNICIPAL_TOPICS (housing, transportation, environment, budget, etc.). Implement agenda item topic classifier. Backfill for all existing items.
• Deterministic stats: compute_legislative_stats() (motion success, dissent rate, co-sponsorship), compute_speaking_stats() (segments/meeting, HHI, question rate), compute_alignment_by_topic(), detect_key_votes() (score 0-5).
• AI profile generation: generate_position_summaries() (top 5 topics), detect_notable_moments(), generate_summary() (2-3 sentences), generate_short_bio() (one-liner).
• Pipeline integration: Add --regenerate-profiles flag. Hook into post-ingestion. Incremental updates only for councillors with new data.
• Profile page redesign: At-a-glance stat cards, AI summary, expandable policy position cards with stance indicators, key votes highlight reel, topic-scoped alignment selector, speaking profile charts.
• RAG integration: Wire get_person_activity tool to read from person_profiles for quick stats.

Profile Page Layout (top to bottom)

1. Header: Name, role, photo, short_bio, tenure
2. At a Glance: Attendance rate (trend), voting record pie, motion success rate, dissent rate, speaking rank
3. AI Summary: 2-3 sentence overview with timestamp and meeting count
4. Policy Positions: Expandable cards per topic with stance indicator, summary, key quotes, key votes
5. Key Votes: Top 5-10 most noteworthy votes with context
6. Voting Alignment: Overall + topic-specific + time-series sparklines + bloc detection
7. Speaking Profile: Topic distribution chart, role breakdown, debate engagement
8. Electoral History (existing, moved lower)
9. Full Records (moved to subpages)

Key Vote Detection Criteria (algorithmic, not AI)

1. Councillor was in the minority
2. Motion was close (passed/failed by 1-2 votes)
3. Councillor broke from usual allies
4. Councillor moved or seconded the motion
5. Motion was on one of their top topics

[kpeatt]

Partially shipped in v1.1

What was built:

• v1.1 Phase 9 (AI Profiling & Comparison, 4 plans, 2026-02-18): Basic person profile pages with voting history, statements, electoral history, attendance stats

What remains (see new issues):

• Topic taxonomy and agenda item classifier
• AI-generated profile summaries and position analysis
• Key vote detection algorithm
• Voting alignment by topic and bloc detection
• Speaking pattern analysis
• Profile page redesign with at-a-glance cards

Closing this umbrella issue. Remaining items tracked as separate focused issues.