Skip to content

Latest commit

 

History

History
130 lines (97 loc) · 3.58 KB

LEARNING_QUICKSTART.md

File metadata and controls

130 lines (97 loc) · 3.58 KB

Quick Start: AI Learning System

What Was Fixed

Your AI wasn't learning because the learning infrastructure existed but wasn't connected to the web UI. I've now integrated it.

Changes Made

1. Enhanced Query Function (web_app.py)

  • ✅ Now generates actual answers (not just retrieval)
  • ✅ Injects learned knowledge from previous interactions
  • ✅ Uses ContextInjector to retrieve relevant facts

2. Added Feedback UI

  • ✅ 👍 Thumbs up button (rating: 0.9)
  • ✅ 👎 Thumbs down button (rating: -0.5)
  • ✅ ⭐ 1-5 star rating slider
  • ✅ 💡 Correction text area for fixing wrong answers

3. Implemented Learning Pipeline

  • ✅ submit_feedback() - Processes user ratings
  • ✅ FeedbackLearner - Extracts facts from good answers
  • ✅ Stores reasoning chains and query refinements
  • ✅ Updates confidence scores based on usage

4. Added Knowledge Visibility

  • ✅ "View Learned Facts" button
  • ✅ "View Query Patterns" button
  • ✅ Shows confidence, usage count, success rate

How to Use

Step 1: Start the App

streamlit run web_app.py

Step 2: Ingest Documents

  1. Go to "Knowledge Base" tab
  2. Enter path to your documentation folder
  3. Click "Ingest Documents"

Step 3: Ask Questions

  1. Type your question in the query box
  2. Click "Search"
  3. AI generates an answer using docs + learned knowledge

Step 4: Provide Feedback

  • Click 👍 if the answer is helpful
  • Click 👎 if it's not helpful
  • Or use the slider for custom rating
  • Provide correction if the answer is wrong

Step 5: See Learning Happen

  • Expand "🧠 View Learned Knowledge"
  • Click "View Learned Facts"
  • See what the AI has extracted and learned!

Example

First Query:

Q: "How do I set Spark executor memory?"
A: "Use spark.executor.memory=4g in your config..."
👍 (User clicks helpful)

→ AI learns: "spark.executor.memory controls executor heap size"

Second Query (Later):

Q: "What's the Spark memory config?"
A: [Uses learned fact from before]
   "Based on previous learning: spark.executor.memory 
   controls the heap size of executors..."

→ Answer is better because it remembers!

Verification

Check if learning is working:

-- Connect to your PostgreSQL database
psql -h localhost -U omniscient -d omniscient

-- Check learned facts
SELECT topic, fact, confidence, usage_count 
FROM learned_facts 
ORDER BY confidence DESC;

-- Check feedback
SELECT question, rating, feedback_type, created_at 
FROM user_feedback 
ORDER BY created_at DESC 
LIMIT 10;

What the AI Learns

  1. Facts - Extracted from highly-rated answers (rating ≥ 0.6)
  2. Reasoning Chains - Step-by-step logic that worked
  3. Query Refinements - Better ways to phrase questions
  4. Corrections - User-provided fixes for wrong answers

Benefits

  • ✅ Continuous Improvement - Gets smarter with every interaction
  • ✅ Domain Expertise - Learns your project-specific knowledge
  • ✅ Error Correction - Users can fix mistakes
  • ✅ Persistent - Knowledge survives restarts
  • ✅ Transparent - See what was learned

Full Documentation

See LEARNING_SYSTEM.md for complete details on:

  • Architecture and data flow
  • Database schema
  • Configuration options
  • Advanced features
  • Troubleshooting

Summary

The Problem: Your AI had a sophisticated learning system built in, but it wasn't being used by the web app.

The Solution: I integrated the existing learning modules (KnowledgeMemory, FeedbackLearner, ContextInjector) into the web UI.

The Result: Your AI now learns from user feedback and improves over time! 🎉