Skip to main content

Let us choose the stack

· 22 min read
Jean-Noël Schilling
Jean-Noël Schilling
Locki one / french maintainer

Project Context Update (Mid-to-Late January 2026)​

Project: Locki / Ò Capistaine (audierne2026) - AI-powered civic transparency & participatory democracy platform for Audierne 2026 local elections (France) Core Mission: Build a neutral, source-based RAG chatbot to answer citizen questions about 4 municipal programs, automate contribution validation against charter rules, crawl/process municipal data (150+ links, 4,000+ PDFs), and showcase Opik integration for the "Commit to Change" Hackathon. Critical Timeline:

  • Contributions deadline: ~January 31, 2026
  • Hackathon prototype delivery: ~mid-February 2026 (4-week sprint)
  • Election period: ~March 15-22, 2026

Team Status & Availability​

Active Members:

  • [jnxmas] aka Johnny Christmass (@jnxmas): Project lead, architecture, N8N deployment, Firecrawl testing, documentation, political outreach

  • Meher/Gurmukher/Guru (@GurmeherSingh): ML Engineer, RAG agent + Opik integration (recently arrived US East Coast, jet-lagged)

  • Victor (@zcbtvag): Backend Python, Firecrawl pipeline (intensive work Mon-Wed, unavailable Thu-Sun for France travel, returns following week)

    Team Experience Alignment:

  • ✅ Shared expertise: Streamlit, VectorStore, Python, Firecrawl, Gemini, Opik

  • ✅ Guru's strengths: NOMIC embeddings (Sentence Transformer), Azure, Ollama

  • ❌ Learning curve: N8N (new for Guru + Victor)

Political & Community Engagement Progress​

Achievements:

  • Outgoing mayor: Positive signal at New Year's event - mentioned "new paradigm in politics"
  • Email follow-up: Mayor acknowledged project email (10+ days prior), interested in discussing Audierne2026
  • Electoral lists: 3 of 4 lists contacted, very positive responses
  • Remaining: 1 list still to contact (before Jan 31) Current Challenge:
  • Low citizen engagement: Contributions/discussions system seeing minimal activity ("it's a bit too much right now")
  • Content workflow: Issues validated → transferred to discussions → posted on website after community review
  • Validation stages: Charter compliance + Contextualization (innovative agent finding solutions) Strategic Shift: Team needs to move from outreach to coding and delivery.

Technical Architecture - Stack Decisions Finalized​

Layer Structure (Bottom-Up ETL Pattern)​

1. External Services Layer

  • Firecrawl API: Primary web scraping/crawling service
  • Status: Victor progressing, PR open, completion target Wednesday (before travel)
  • Blocker: Pagination settings need tuning (tested with max_pages=10, returned only 1 page)
  • Priority source: Délibérations du conseil municipal (municipal council deliberations) 2. Data Access Layer
  • Vector Store: Confirmed standard VectorStore approach (all team members familiar)
  • Embeddings Strategy:
    • Primary: NOMIC model (Sentence Transformer) - Guru has experience
    • Backup/Failsafe: Mistral Studio - French language optimization + potential local sponsorship
  • Crawl Tracking Database: SQL Lite/SQL Model chosen
    • Metadata: Crawl status, RAG inclusion, content category
    • SQL Model benefits: Type safety (Python → SQL columns), FastAPI/Pydantic compatibility, PostgreSQL support
    • Owner: [jnxmas] will build during Victor's absence 3. Business Logic Layer Agent 1: Charter Validation Agent (PRIORITY)
  • Purpose: Auto-validate citizen contributions against charter rules
  • Checks: Respectful language, no personal naming, constructive content
  • Functionality: Binary (green/red) conformance + auto-correct miscategorized submissions (7 categories)
  • Tech Stack: Gemini (primary), Ollama (testing), Opik evaluation/tracing
  • Workflow: Reads audierne2026 repo issues → labels/updates in ocapistaine repo
  • Owner: Guru (@GurmeherSingh)
  • Status: Not started (blocked by Opik access + N8N deployment) Agent 2: RAG Agent / Citizen Q&A Chatbot
  • Purpose: Answer citizen questions with neutral, contextual responses, cross-reference municipal programs
  • Requirements: No hallucinations (Opik guardrails), source attribution, French language
  • Approach: Start with available data, iterate as crawling progresses
  • Owner: Guru (@GurmeherSingh)
  • Status: Starting development, update expected in 1-2 days
  • Dependencies: VectorStore populated, Charter agent outputs (optional) Agent 3: Search Agent (CONDITIONAL)
  • Decision: Pending RAG completeness assessment
  • Logic: If RAG sufficient → search agent unnecessary; if RAG insufficient → search agent as fallback
  • Status: On hold pending RAG testing Chat Service (Already Working)
  • Functionality: Multi-turn conversations with thread management (UUID per user)
  • Providers: OpenAI, Ollama (Mistral, DeepSeek tested locally), Perplexity, Gemini
  • Features: System message config, prompt engineering, thread history persistence, strategy selection (Charter mode, Q&A mode, etc.)
  • Opik Integration: Full tracing via workflow_chat class Document Service (To Be Defined)
  • Proposed Purpose:
    • Query documentation/source information
    • Scheduled re-crawling (detect new docs since last crawl)
    • Document categorization + metadata tagging
    • Change detection (compare current vs. new crawls, decide RAG updates)
  • Open Question: How does this differ from Chat Service? (Clarified: Document = search/crawl management; Chat = conversational interface)
  • Status: Requirements to be defined collaboratively 4. Application Layer
  • Streamlit: Primary presentation interface
  • FastAPI: Routing layer
  • Debug Scripts: One-click full stack launch (UV cone + Streamlit app automation)
  • Hot Reload: Code modifications work without restart
  • Future: Redis integration for scheduler + task memory persistence 5. External Orchestration Layer
  • N8N: Self-hosted Docker deployment on Contabo server (viety.loki.io)
  • Status: ~2 hours from completion (as of last meeting)
  • Workflows:
    • Pull issues from audierne2026 GitHub repo
    • Trigger charter validation agent
    • Push validated/labeled issues back to repo
    • Modify Jekyll website content
    • Provide API endpoint for agent testing
  • Multi-project approach: Separate workflows per project on shared dashboard
  • Future: Pull data from Opik Cloud back into app (feedback loop via N8N API) Supporting Infrastructure:
  • APScheduler (cron-based): Time-based tasks (daily/weekly crawl updates, GitHub monitoring, Charter violation checks)
  • Opik: Evaluation, tracing, prompt management (future), hallucination detection, metrics/dashboards
  • Poetry: Dependency management (NOT pip) for library compatibility

Opik Integration Strategy - MAXIMIZE USAGE​

Team Consensus: "Maximize Opik usage at every opportunity" - it's the hackathon focus. Dashboard Setup Decisions:

  • Environment: Start with Opik Dev, migrate to Prod when ready
  • Dashboard Approach: Single unified dashboard to monitor all agents (avoid complexity of separate dashboards per agent)
  • API Keys: Likely one project with multiple keys for per-developer usage tracking
  • Access:
    • [jnxmas] has access to ocapistaine-dev dashboard
    • Need to grant Guru and Victor access
    • Guru has existing Opik account (needs to share user ID/Discord username)
    • Victor's Opik status: Not confirmed Integration Points:
  • Charter validation agent: Trace every decision, verify adherence to charter rules
  • RAG agent: Hallucination prevention, prompt evaluation, guardrails
  • N8N workflows: Plug into Opik for agent activity reporting
  • Chat service: Already using Opik tracing via workflow_chat class
  • Prompt management: Start hardcoded, potentially migrate to Opik prompt library later
  • Evaluation judge: Verify LLM adherence to prompts Team Familiarity:
  • Platform exploration phase: Team has not yet fully explored Opik capabilities
  • Reference materials: Git examples available for implementation patterns
  • Action needed: All team members familiarize with Opik features (dashboards, prompt library, evaluation)

LLM Provider Strategy - Multi-Provider Approach​

Confirmed Providers:

  1. Gemini: Sponsored/free option (primary for Charter agent)
  2. Ollama: Completely free (local testing) - Mistral, DeepSeek tested
  3. Mistral: French language optimization + potential sponsorship target (French govt/local alignment, political risk mitigation)
  4. OpenAI: Already working in Chat service
  5. Perplexity: Available option Strategic Rationale:
  • French language: Critical for citizen-facing content - Mistral chosen for optimization
  • Political optics: Using French LLM provider avoids criticism of not supporting French solutions
  • Cost management: Prioritize free/sponsored options (Gemini, Ollama)
  • Failsafe approach: NOMIC primary + Mistral backup = "two tracks maximum" ([jnxmas]'s preference given 4-week timeline)

Development Workflow & Version Control​

Git Strategy:

  • Protected branch: dev branch - no direct pushes
  • Feature branches: Mandatory for all work
  • PR workflow: Victor successfully resolved upstream branch issue, has open PR ready for review
  • Review process: [jnxmas] reviews and merges into dev Dependency Management:
  • Poetry ONLY: NOT pip - ensures library compatibility
  • Command: poetry add
  • Action: Victor + Guru ensure local environments use Poetry Documentation:
  • Docusaurus blog: All meeting summaries posted to blog/ for async collaboration + AI context (Cursor/Claude code context)
  • GitHub tasks: Self-assignment encouraged, commenting for async feedback
  • Task board: 3 tasks "In Progress" with all three team members assigned

Critical Technical Decisions Summary​

ComponentDecisionOwnerStatus
Vector DBVectorStore (standard)Team✅ Confirmed
EmbeddingsNOMIC primary, Mistral backupGuru✅ Confirmed
OrchestrationN8N (Docker, viety.loki.io)[jnxmas]🟡 ~2h from deployment
Crawl TrackingSQL Lite + SQL Model[jnxmas]🔴 Design phase
LLM ProvidersGemini, Ollama, Mistral, OpenAI, PerplexityTeam✅ Multi-provider
Opik SetupSingle dashboard, Dev → Prod, multiple API keys[jnxmas]🟡 Access pending
OCR Librariespdf2ocr, Tabula, pypdf (test all 3)Victor → [jnxmas]🔴 Blocked on crawling
Prompt ManagementHardcoded → Opik library (future)Guru🔴 Not started
Search AgentConditional on RAG completenessTBD🟡 On hold

Firecrawl Pipeline Progress (Victor's Focus)​

Completed:

  • ✅ Debugged and fixed scraper component
  • ✅ Fixed crawler functionality (operational after "small but critical" config issues)
  • ✅ Open PR ready for review
  • ✅ Resolved upstream branch issue Current Issues:
  • ⚠️ Pagination settings need adjustment (tested max_pages=10, returned only 1 page)
  • ⚠️ Full document retrieval tuning required Victor's Timeline:
  • Mon-Wed (this week): Intensive work, complete Firecrawl pipeline
  • Thu-Sun: Unavailable (travel to France - pre-planned holiday)
  • Following week: Returns and continues contributions
  • Goal: Finish Firecrawl work before Wednesday departure Next Steps:
  1. Victor finalizes additional commits to feature branch
  2. Victor signals when PR ready for final review
  3. [jnxmas] reviews and merges into dev
  4. [jnxmas] tests merged pipeline on additional sources during Victor's absence
  5. Once crawling complete → Apply OCR to PDFs (pdf2ocr, Tabula, pypdf testing)

OCR Pipeline Strategy (Post-Firecrawl)​

Target Documents: Scanned PDFs in municipal archives (~4,000+ documents) Libraries to Test:

  1. pdf2ocr
  2. Tabula
  3. pypdf Approach: Test all 3, compare outputs for quality Owner: Victor (primary), [jnxmas] (testing during Victor's absence) Status: Blocked on Firecrawl crawling completion

Project Background & Context​

Locki Project Evolution:

  • Origin: Started from hackathon 2 years ago
  • Built agents on top of Locki stack
  • Participated in cohorts to improve functionality
  • Previous projects:
    • 3D objects on-chain (paused - too early/complex, slow blockchain)
    • Horse racing betting strategy (15 years expert knowledge → code/data layer, not yet commercialized)
    • Multiverse Six / XRP experience (3D NFTs) [jnxmas]'s Work:
  • Freelancing: Apps, websites, e-commerce development
  • Computer teacher for small children (Thursdays, 2 hours)
  • Locki is primary passion project ("what makes me wake up in the morning")
  • Successfully built Locki mobile app prototype Dev Environment:
  • VS Code with debug scripts for one-click launch
  • Team using Cursor IDE may need to adapt scripts
  • Victor considering switch to VS Code for easier workflow

Open Questions & Decisions Needed​

For Guru (Flagged for Response):​

  1. ✅ Vector store preference: VectorStore confirmed, NOMIC embeddings confirmed
  2. ⏳ Opik user ID/Discord username: Needed for dashboard access grant
  3. ⏳ Stack alignment review: Read architecture blog post in repo, suggest improvements
  4. ⏳ Charter agent design: Begin development, provide update in 1-2 days

For Victor:​

  1. ⏳ Opik account status: Confirm if account exists or needs creation
  2. ⏳ Firecrawl PR readiness: Signal when ready for final review
  3. ⏳ Poetry environment: Verify local setup uses Poetry (not pip)

For [jnxmas]:​

  1. ⏳ N8N deployment: Complete final ~2 hours of setup
  2. ⏳ Opik dashboard access: Invite Guru + Victor (need usernames)
  3. ⏳ Document Service requirements: Define scope and distinction from Chat Service
  4. ⏳ Crawl scheduling: Decide frequency (daily vs. weekly) for production

For Team Discussion:​

  1. ⏳ Search Agent decision: Revisit after RAG testing complete
  2. ⏳ Citizen engagement UX: Consider simplifying contribution system if low engagement persists
  3. ⏳ Effort attribution: Finalize "all ores tracking template" for fair team contribution recognition
  4. ⏳ Prompt management migration: Timeline for moving hardcoded prompts to Opik library

Comprehensive Task List​

URGENT (Next 48 Hours)​

Task 1: Complete N8N Deployment & GitHub Integration

  • Owner: [jnxmas] (@[jnxmas])
  • Description: Finish final ~2 hours of N8N Docker deployment on viety.loki.io, configure workflows to pull issues from audierne2026 repo, trigger charter validation agent, push validated/labeled issues back, provide API endpoint for testing
  • Deadline: Within 48 hours (before Guru needs to test agent)
  • Dependencies: None (infrastructure task)
  • Success Criteria: N8N accessible at viety.loki.io, test workflow successfully pulls/pushes GitHub issue, API endpoint documented Task 2: Set Up Opik Dev Environment & Team Access
  • Owner: [jnxmas] (@[jnxmas])
  • Description: Create/configure Ocapistaine Dev Opik project, generate separate API keys for [jnxmas]/Guru/Victor, grant dashboard access to all team members, document key assignment and usage tracking approach
  • Deadline: Before Guru starts agent development (next 24 hours)
  • Dependencies: Guru's Opik user ID/Discord username, Victor's Opik account confirmation
  • Success Criteria: All team members can access dashboard, separate keys functional, usage trackable per developer Task 3: Complete Firecrawl Pipeline Implementation
  • Owner: Victor (@zcbtvag)
  • Description: Finalize additional commits to Firecrawl feature branch, fix pagination settings to ensure full document retrieval, test with multiple sources, signal when PR ready for final review
  • Deadline: Wednesday (before France travel)
  • Dependencies: Firecrawl API keys, crawler pagination tuning
  • Success Criteria: PR merged without conflicts, Firecrawl successfully crawls full test dataset, documented in code Task 4: Build Charter Validation Agent - Core Logic
  • Owner: Guru (@GurmeherSingh)
  • Description: Develop charter validation agent that reads citizen contribution issues from audierne2026 repo, validates against charter rules (respectful, no naming, constructive), auto-corrects category misclassifications (7 categories), returns binary validation + corrected category, integrates Opik evaluation at every decision point
  • Deadline: Update in 1-2 days, full prototype ASAP (foundation for other workflows)
  • Dependencies: Access to Opik dashboard, audierne2026 repo read access, N8N API endpoint
  • Success Criteria: Agent correctly validates 95%+ test cases, Opik traces show decision transparency, category corrections work

HIGH PRIORITY (This Week)​

Task 5: Build RAG Agent with Opik Integration

  • Owner: Guru (@GurmeherSingh)
  • Description: Develop core RAG agent for citizen Q&A, integrate NOMIC embeddings (primary) + Mistral (backup), implement Opik for evaluation/tracing/hallucination prevention, generate neutral contextual responses with source attribution, cross-reference municipal program data
  • Deadline: Update in 1-2 days, working prototype by end of week (Jan 30-31)
  • Dependencies: VectorStore setup, crawled data (start with available), Opik access
  • Success Criteria: Functional RAG agent with Opik tracing active, no hallucinations in test cases, proper French language responses Task 6: Firecrawl PR Review & Merge
  • Owner: Victor (@zcbtvag) + [jnxmas] (review)
  • Description: Victor signals PR ready, [jnxmas] reviews and merges into dev branch, test merged pipeline functionality
  • Deadline: End of week (allow Victor time for improvements)
  • Dependencies: Task 3 completion
  • Success Criteria: PR merged, Firecrawl pipeline functional on test sources Task 7: Test Firecrawl on Additional Sources & Build Crawl Tracking DB
  • Owner: [jnxmas] (@[jnxmas])
  • Description: Intensively test Victor's Firecrawl code on ocapistaine repo PDFs/websites and other sources beyond current scope, begin SQL Lite/SQL Model database for tracking crawled content with metadata (crawl status, RAG inclusion, content category)
  • Deadline: Sunday evening (status update)
  • Dependencies: Task 6 completion (merged Firecrawl code)
  • Success Criteria: Multiple sources tested successfully, tracking database prototype with crawl status and categorization functional Task 8: Explore Opik Platform Features
  • Owner: All team members
  • Description: Familiarize with Opik capabilities, dashboards, prompt library, evaluation features, tracing, metrics, experiment tracking
  • Deadline: Next 2-3 days (parallel with development)
  • Dependencies: Opik access granted (Task 2)
  • Success Criteria: Team comfortable with platform, ready to maximize hackathon showcase, documented learnings Task 9: Poetry Dependency Management Setup
  • Owner: Victor (@zcbtvag) + Guru (@GurmeherSingh)
  • Description: Ensure local dev environments use Poetry (not pip) for all library additions, verify compatible with current workflows
  • Deadline: Before next library additions
  • Dependencies: Poetry installed locally
  • Success Criteria: No pip-related dependency conflicts, pyproject.toml up to date, team using poetry add exclusively

MEDIUM PRIORITY (Next Week)​

Task 10: Test OCR Libraries on Scanned PDFs

  • Owner: Victor (@zcbtvag) → [jnxmas] (if Victor unavailable)
  • Description: Test pdf2ocr, Tabula, pypdf on sample scanned PDFs from municipal archives, compare outputs for quality, select best library/combination for production
  • Deadline: Immediately after crawling complete (or during Victor's absence)
  • Dependencies: Task 3/6 completion (crawling functional)
  • Success Criteria: 3 libraries tested, quality comparison documented, production OCR library selected Task 11: Architecture Documentation Review
  • Owner: Guru (@GurmeherSingh) + Victor (@zcbtvag)
  • Description: Read architecture blog post [jnxmas] created (in ocapistaine repo), suggest additions specifically around NOMIC embedding implementation details, RAG service integration points, any conflicts/improvements from experience
  • Deadline: Next 2-3 days (before architecture freeze)
  • Dependencies: Access to blog in repo
  • Success Criteria: Comments/suggestions posted, team aligned on architecture Task 12: Define Document Service Requirements
  • Owner: [jnxmas] + Guru + Victor
  • Description: Collaboratively define Document Service scope: query documentation/source info, scheduled re-crawling (detect new docs), document categorization + metadata tagging, change detection logic (compare current vs. new crawls), decide if separate DB needed, clarify distinction from Chat Service
  • Deadline: Next sync meeting
  • Dependencies: Vector store decision (confirmed), crawl tracking DB design (Task 7)
  • Success Criteria: Document Service requirements documented, team aligned, ready for implementation Task 13: Finish All Ores Tracking Template
  • Owner: [jnxmas] (@[jnxmas])
  • Description: Complete effort attribution tracking system for fair team contribution recognition, share with team for feedback
  • Deadline: ASAP (ongoing discussion on Discord)
  • Dependencies: None
  • Success Criteria: Template finalized, shared on Discord, team feedback incorporated Task 14: Follow Up with Remaining Electoral List
  • Owner: [jnxmas] (@[jnxmas])
  • Description: Contact 4th electoral list for outreach, gauge interest in Audierne2026 project
  • Deadline: Before January 31 (contributions deadline)
  • Dependencies: None
  • Success Criteria: All 4 lists contacted, responses documented

FUTURE / CONDITIONAL​

Task 15: Search Agent Design & Implementation (CONDITIONAL)

  • Owner: TBD (likely Guru)
  • Description: IF RAG agent proves insufficient during testing, design and implement Search Agent as fallback for citizen queries
  • Deadline: Week 2 of sprint (after RAG testing complete)
  • Dependencies: Task 5 completion (RAG agent functional), RAG sufficiency evaluation
  • Success Criteria: Decision documented (Search Agent needed: yes/no), if yes → design document with Opik integration plan Task 16: Migrate Prompts to Opik Library (FUTURE)
  • Owner: Guru (@GurmeherSingh)
  • Description: After initial hardcoded prompts working, migrate to Opik prompt library for centralized management, versioning, experimentation
  • Deadline: Post-hackathon or if time permits in sprint
  • Dependencies: Opik platform familiarity (Task 8), working agents with hardcoded prompts
  • Success Criteria: Prompts managed in Opik, agents use library-based prompts, versioning functional Task 17: Redis Integration for Scheduler + Task Memory
  • Owner: [jnxmas] (@[jnxmas])
  • Description: Integrate Redis for APScheduler persistence and task memory across restarts
  • Deadline: Post-initial prototype (when scaling/production readiness needed)
  • Dependencies: Working scheduler tasks (crawl updates, GitHub monitoring)
  • Success Criteria: Scheduler tasks persist across app restarts, task memory functional Task 18: Citizen Q&A RAG Agent - Planning (After Charter Agent)
  • Owner: Guru (@GurmeherSingh)
  • Description: After charter agent working, design second RAG agent iteration for citizen questions: reads validated contributions from charter agent, generates neutral contextual responses, cross-references municipal program data, no hallucinations (Opik guardrails)
  • Deadline: Week 2 of sprint (after charter agent complete)
  • Dependencies: Charter agent functional (Task 4), N8N workflows ready, VectorStore populated
  • Success Criteria: Design document with Opik integration plan, data flow diagram

Risk Assessment & Mitigations​

RiskImpactMitigationOwnerStatus
Guru missing from stack discussionMediumRecord meetings, request written feedback on tasks, schedule follow-up calls[jnxmas]🟢 Mitigated (recordings shared)
Victor unavailable Thu-SunMediumVictor completes Firecrawl by Wed, [jnxmas] tests during absenceVictor + [jnxmas]🟢 Planned
Crawler pagination not returning full resultsHighVictor debugging this week; if blocked, escalate to teamVictor🟡 In progress
N8N deployment blocking agent testingHigh[jnxmas] prioritizes ~2h completion ASAP[jnxmas]🟡 In progress
Opik access blocking Guru's developmentHigh[jnxmas] grants access within 24h (need Guru's username)[jnxmas]🟡 In progress
Vector store decision blocking RAGMediumVectorStore + NOMIC confirmed; fallback = Mistral StudioGuru + [jnxmas]🟢 Resolved
OCR quality unknown for scanned PDFsMediumTest 3 libraries (pdf2ocr, Tabula, pypdf), compare outputsVictor → [jnxmas]🟡 Planned
4-week sprint timeline pressureHighPrioritize working prototype over perfection, ruthless task board management, limit to "two tracks" (NOMIC + Mistral)All🟡 Ongoing
First agent not live before contributions deadline (~Feb 1)CriticalFast-track Charter agent development this week, daily standupsGuru + [jnxmas]🔴 Active risk
Low citizen engagement with contributionsMediumConsider UX simplification, focus on chatbot quality over contribution volume[jnxmas]🟡 Monitoring
Team still in ideation phase despite deadline pressureMediumShift to execution mode immediately, lock architecture decisionsAll🟡 Transitioning
French language hallucinations/bias in chatbotCriticalMistral backup, heavy Opik guardrails, manual testing in FrenchGuru + [jnxmas]🟡 Design phase
Hackathon Opik showcase insufficientMedium"Maximize Opik at every opportunity", document integration patterns, trace everythingGuru🟡 In progress

Alignment Confirmations & Strategic Notes​

✅ Strong Alignments:

  • Opik-first approach aligns perfectly with hackathon goals
  • Multi-provider LLM strategy provides resilience and cost optimization
  • French language focus (Mistral) addresses political optics and user needs
  • Modular ETL architecture enables swapping components without rewrites
  • Team expertise (NOMIC, Streamlit, VectorStore) matches stack decisions
  • Task ownership clear and distributed well (Guru = agents, Victor = crawling, [jnxmas] = infrastructure/orchestration) ⚠️ Potential Misalignments:
  • Timeline vs. scope: 4 weeks for full pipeline (crawling + OCR + RAG + Charter agent + chatbot + Opik showcase) is aggressive
    • Mitigation: Limit to "two tracks" (NOMIC primary, Mistral backup), prioritize Charter agent + basic RAG over perfect Search Agent
  • Citizen engagement vs. effort: Low contribution volume may not justify complex validation workflow
    • Mitigation: Focus on chatbot quality (broader citizen reach) over contribution automation
  • Team availability gaps: Victor's Thu-Sun absence, Guru's jet lag, [jnxmas]'s teaching commitment
    • Mitigation: Strong async collaboration (recordings, blog summaries, GitHub comments), clear handoffs 🔍 Clarifications Needed:
  • Document Service vs. Chat Service: Distinction still fuzzy (addressed as: Document = search/crawl management; Chat = conversational interface, but needs formal requirements doc)
  • Search Agent necessity: Conditional decision pending RAG testing (good pragmatic approach)
  • Opik dashboard structure: Single unified dashboard confirmed, but team needs to explore platform features to maximize value 📈 Recommended Focus Areas:
  1. This week: Charter agent + RAG agent + Firecrawl completion + Opik access = critical path
  2. Next week: OCR testing + Document Service definition + crawl tracking DB = data pipeline completion
  3. Week 3-4: Integration, testing, French language validation, Opik showcase polish, citizen-facing chatbot MVP

Documentation & Knowledge Management​

Current Practices (Strong):

  • ✅ Docusaurus blog: Meeting summaries posted to blog/ for async collaboration
  • ✅ GitHub tasks: Self-assignment, commenting for feedback
  • ✅ Task board: 3 tasks "In Progress" with assignments
  • ✅ Architecture blog post: Created for team review + AI context (Cursor/Claude)
  • ✅ Meeting recordings: Shared for async team members (Guru)
  • ✅ "Tell what we do and do what we tell": Commitment tracking across conversations Improvements Suggested:
  • 📝 Document Opik integration patterns as you build (hackathon showcase value)
  • 📝 Create crawl tracking DB schema diagram (avoid ambiguity)
  • 📝 Formalize Document Service requirements doc (per Task 12)
  • 📝 Keep citizen-facing deliverable timeline visible (dashboard/Gantt chart)
  • 📝 Document OCR library comparison results (Task 10)

Next Sync Meeting Agenda (Suggested)​

  1. N8N + Opik access: Confirm both unblocked ([jnxmas] updates)
  2. Firecrawl PR status: Review and merge if ready (Victor + [jnxmas])
  3. Charter agent progress: Demo initial prototype, discuss Opik traces (Guru)
  4. RAG agent update: Share approach, blockers, timeline (Guru)
  5. Crawl tracking DB: Review schema design ([jnxmas])
  6. Document Service requirements: Collaborative definition session (All)
  7. Task board review: Close completed tasks, reprioritize open items (All)
  8. Opik showcase strategy: What metrics/traces will impress hackathon judges? (All)

Status Dashboard​

Overall Progress:

  • 🟢 Architecture: Finalized (VectorStore, NOMIC, Mistral, N8N, Opik strategy confirmed)
  • 🟡 Firecrawl Pipeline: Victor progressing, PR open, completion target Wed
  • 🟡 N8N Deployment: ~2 hours from completion (infrastructure blocker)
  • 🟡 Opik Setup: Account exists, team access pending (need usernames)
  • 🔴 Charter Agent: Not started (blocked by Opik access + N8N)
  • 🟡 RAG Agent: Starting development, update expected 1-2 days
  • 🔴 OCR Pipeline: Blocked on crawling completion
  • 🔴 Crawl Tracking DB: Design phase
  • 🟢 Team Coordination: Strong communication, clear roles, async collaboration working
  • 🟡 Political Outreach: 3/4 lists contacted, positive signals
  • 🔴 Citizen Engagement: Low contribution volume (UX consideration needed)
  • 🟢 Documentation: Blog summaries active, recordings shared, task tracking functional Open High-Priority Tasks (Top 5):
  1. 🔴 URGENT: Complete N8N deployment ([jnxmas], <48h) - blocks agent testing
  2. 🔴 URGENT: Set up Opik team access ([jnxmas], <24h) - blocks Guru's development
  3. 🔴 URGENT: Complete Firecrawl pipeline (Victor, by Wed) - data ingestion dependency
  4. 🔴 HIGH: Build Charter validation agent (Guru, update 1-2 days) - critical path for hackathon
  5. 🔴 HIGH: Build RAG agent with Opik (Guru, update 1-2 days) - citizen-facing deliverable Next Milestones:
  • Week 1 Goal (by ~Jan 30-31): Charter validation agent functional with Opik tracing + Firecrawl pipeline merged + N8N orchestrating GitHub workflows + RAG agent prototype
  • Immediate Blockers: N8N deployment (<48h), Opik access (<24h), Firecrawl pagination tuning (<3 days)
  • Hackathon Prototype Delivery: ~mid-February 2026 (2.5 weeks remaining from late Jan)
  • Critical Path: Crawling → OCR → Vector DB → RAG + Opik → Chatbot MVP → French language validation Team Availability This Week:
  • Guru: Available now (jet-lagged but active), development starting
  • Victor: Intensive Mon-Wed, unavailable Thu-Sun (France travel)
  • [jnxmas]: Full availability, prioritizing N8N/Opik/testing during Victor's absence Risk Level: 🟡 MEDIUM-HIGH - Timeline tight, first agent not yet live despite approaching contributions deadline, but team aligned and architecture solid. Immediate focus on unblocking N8N + Opik access will shift to 🟢 GREEN if completed within 48h.

Action Summary - What Happens Next​

Immediate (Next 24 Hours):

  1. [jnxmas] completes N8N deployment (~2 hours)
  2. [jnxmas] sets up Opik team access (needs Guru + Victor usernames via Discord)
  3. Guru shares Opik user ID/Discord username
  4. Victor continues Firecrawl pagination debugging
  5. Team explores Opik platform features once access granted

This Week (Next 3-7 Days):

  1. Victor completes Firecrawl PR by Wednesday
  2. [jnxmas] reviews and merges Firecrawl PR
  3. Guru develops Charter validation agent, provides update in 1-2 days
  4. Guru develops RAG agent prototype, provides update in 1-2 days
  5. [jnxmas] tests Firecrawl on additional sources during Victor's absence
  6. [jnxmas] begins crawl tracking DB with SQL Model
  7. Guru + Victor review architecture blog post, suggest improvements
  8. All team members familiarize with Opik capabilities

Next Week (Week 2 of Sprint):

  1. OCR library testing begins (Victor returns, or [jnxmas] covers)
  2. Document Service requirements defined collaboratively
  3. RAG agent fully functional with Opik guardrails
  4. Charter agent deployed to N8N workflows
  5. Crawl tracking DB prototype operational
  6. Architecture frozen, team shifts to integration phase

Week 3-4 (Final Sprint):

  1. Full pipeline integration testing
  2. French language validation and bias checking
  3. Citizen-facing chatbot MVP deployment
  4. Opik showcase polish (metrics, traces, dashboards, experiment tracking)
  5. Hackathon submission preparation
  6. Final outreach to 4th electoral list
  7. Documentation finalization

Confidence Level: 🟡 CAUTIOUSLY OPTIMISTIC - Strong team, solid architecture, clear tasks, but timeline aggressive and first agent not yet live. Success depends on unblocking N8N + Opik access within 48h and Charter agent prototype within 1 week. Victor's Wed deadline and Guru's jet lag are manageable with current plans. Recommendation: Daily async standups (Discord) this week to track Charter agent + Firecrawl progress. Consider lightweight demo/review session Friday to validate week 1 progress before weekend.