Hugging Face's logo

Spaces:
Abdalkaderdev
/
ORA
Sleeping

App Files Community
ORA / task.md
Abdalkaderdev's picture
Abdalkaderdev
Initial ORA deployment
5e0532d
preview code
|
raw
history blame contribute delete
3.25 kB

ORA System Review Task List

  • Initial System Review

    • Explore codebase structure
    • Identify key configuration and entry points
    • Analyze Services
      • Review orchestrator.py
      • Review memory.py
      • Review guardrails.py
      • Review intent.py & emotion.py
      • Review spiritual practices (prayer/discernment/lectio)
      • Review infrastructure (main.py, config.py)
      • Review peripheral services (llm.py, telemetry.py)
    • Synthesize Review Report
      • Document Architecture/Data Flow
      • Identify Missing Persistent Layers
    • Define "Not to build" list
    • Create One-week MVP plan

  • Implement MVP Core Services

    • Implement Audit Persistence (File logging)
    • Implement Journal Service (SQLite)
    • Implement User Profiles (SQLite)
    • Connect Bible RAG (Vector Search)

  • Phase 2: Advanced Reasoning & Study

    • Research RLM Blog Post
    • Implement RLM (for Deep Study)
      • Design Python REPL Sandbox
      • Implement RLMService wrapper
    • Integrate TRM (for Logic Solvers)
      • Define Theological Logic dataset schema
    • Phase 2 Architecture Upgrades
      • Migrate Storage to LanceDB
      • Ingest bible-nlp/biblenlp-corpus
    • Evaluate Pre-trained Bible Models
      • Test Christian-Bible-Expert-12B
      • Create Gabriel-8x7B via Ollama
    • Setup Unsloth Fine-tuning Pipeline
    • Start "Reasoning Trace" Dataset Collection

  • Phase 3: Frontend Mastery & Integration

    • Migrate soapbox-demo to root as frontend
    • Connect ChatWidget to ORA Backend (Port 6000)
    • Implement Agent Handoff Visualizations
    • Create Handover Prompt for Claude

  • [/] Phase 4: Model Optimization & Fine-tuning

    • [/] Benchmark gabriel-8x7b baseline capabilities
    • Curate theological dataset for specialist fine-tuning
    • Perform LoRA fine-tuning on gabriel-8x7b using Unsloth
    • Evaluate fine-tuned model vs baseline
    • Deploy optimized reasoning model to production

  • General Research Phase

    • Research Local Vector Databases (Chroma/LanceDB)
    • Research Secure REPL Sandboxes (for RLM)
    • Research Fine-tuning Logic for Small Models
    • Evaluate new Bible Datasets (Helsinki, BibleNLP, etc.)
    • Evaluate Pre-trained Bible Models (Gabriel, Christian-Bible-Expert)