How prahalad-ship-it/knowable- works
Episteme AI is a confidence-aware question-answering product: users submit a question, the backend obtains a model response, and the interface presents calibrated text, confidence indicators, gaps, and resolution suggestions. The README also describes asynchronous edge-case auditing and uncertainty logging into a fine-tuning flywheel; the sampled API and triage code support background capture and flywheel handoff, but do not establish a separate audit-worker implementation. Research and calibration scripts form a supporting empirical-evaluation branch. Some frontend stream phases are timed UI states, not evidence of token streaming.
Main entry points: App.jsx Application state · index.py Flask API and orchestration · episteme_empirical_results.csv Empirical results · episteme_research.py Inference benchmark
Components 18
User experience
- Application state
frontend/src/App.jsx - Question input
frontend/src/components/QueryInput.jsx - Audit phase display
frontend/src/components/LiveStream.jsx - Calibrated results UI
frontend/src/components/ResponseCard.jsx - Scenario examples
frontend/src/mockData.js - HTTP client
frontend/src/api.js
Inference and confidence
- Flask API and orchestration
api/index.py - NVIDIA-compatible model
- Confidence and triage processor
triage_protocol_processor.py
Uncertainty flywheel
- Flywheel JSONL data
data
Empirical evaluation
- Inference benchmark
episteme_research.py - Benchmark questions
TruthfulQA.csv - Empirical results
episteme_empirical_results.csv - Uncertainty analysis
analyze_uncertainty_patterns.py - Dashboard metrics
episteme_dashboard_metrics.py - Error analysis
error_analysis.py
Other components
- Question asker
- Flywheel webhook
Connections 19
- Question asker → Question input: enters question
frontend/src/components/QueryInput.jsx - Question input → Application state: submits text
frontend/src/components/QueryInput.jsx - Application state → HTTP client: requests answer
frontend/src/App.jsx - Application state → Question input: renders
frontend/src/App.jsx - Application state → Audit phase display: renders phases
frontend/src/App.jsx - Application state → Calibrated results UI: renders result
frontend/src/App.jsx - Application state → Scenario examples: loads examples
frontend/src/App.jsx - HTTP client → Flask API and orchestration: POSTs question
frontend/src/api.js - Flask API and orchestration → HTTP client: returns payload
README.md - Flask API and orchestration → NVIDIA-compatible model: requests completion
api/index.py - Flask API and orchestration → Confidence and triage processor: uses processor
api/index.py - Flask API and orchestration → Flywheel JSONL data: captures locally
- Flask API and orchestration → Flywheel webhook: hands off asynchronously
- Inference benchmark → Benchmark questions: reads benchmark
episteme_research.py - Inference benchmark → NVIDIA-compatible model: requests completions
episteme_research.py - Inference benchmark → Empirical results: records evaluations
episteme_research.py - Uncertainty analysis → Empirical results: analyzes results
- Dashboard metrics → Empirical results: summarizes metrics
- Error analysis → Empirical results: examines errors