feat: enhance changelog and implement SmartLearner for pattern extraction

CHANGELOG.md

@@ -1,38 +1,59 @@
-# Changelog
+# BuddAI Development Changelog
 
-## [3.2.0] - 2025-12-29
+## Version 3.2 - Self-Learning & Optimization System
 
-### Added
+### 📊 Phase 1: Data Collection
 
-- WebSocket streaming for real-time token-by-token responses
-- Multi-user support with session isolation
-- Connection pooling for Ollama requests (10 connection pool)
-- File upload validation (50MB limit, type checking)
-- Zip slip protection for malicious archives
-- Filename sanitization
-- Type hints throughout codebase
-- Enhanced iOS PWA support
+Implemented comprehensive data logging to track user interactions and code quality.
 
-### Security
+- **Correction Logging**: Added `save_correction()` to store original vs. corrected code with user reasoning.
+- **Compilation Logs**: Added `log_compilation_result()` to track hardware-specific compilation success rates.
+- **Feedback System**: Enhanced `record_feedback()` to support comments and trigger failure analysis on negative feedback.
+- **Database Updates**: Added tables for `corrections`, `compilation_log`, `feedback` (enhanced), and `code_rules`.
 
-- File size limits enforced (50MB)
-- Magic number validation for ZIP files
-- Path traversal prevention in zip extraction
-- Maximum upload file count (10 files)
-- Sanitized filenames to prevent path injection
+### 🔬 Phase 2: Pattern Extraction
 
-### Performance
+Added intelligence to learn from the collected data.
 
-- Connection pooling reduces latency by ~30%
-- WebSocket streaming improves perceived response time
-- Per-user instance management
+- **Smart Learner**: Created `SmartLearner` class to diff code and extract patterns (e.g., `analogWrite` -> `ledcWrite`).
+- **Hardware Profiles**: Created `HardwareProfile` class to manage hardware-specific syntax (ESP32 vs Arduino).
+- **Rule Storage**: Learned patterns are stored in `code_rules` with confidence scores.
+- **Prompt Injection**: `build_enhanced_prompt()` now dynamically injects high-confidence rules into the system prompt.
 
-### Fixed
+### ✅ Phase 3: Validation
 
-- Session isolation bug (users can no longer see each other's data)
-- Connection leak in Ollama requests
-- Memory growth in long-running server instances
+Implemented pre-flight checks to ensure code quality before display.
 
-## [3.1.0] - 2025-12-29
+- **Code Validator**: Created `CodeValidator` class to check for:
+  - ESP32 PWM usage (ledcWrite enforcement).
+  - Blocking delays in motor code.
+  - Missing safety timeouts.
+- **Auto-Fix**: The system can now automatically patch critical errors (like incorrect PWM calls) before showing code to the user.
+- **Hardware Detection**: Automatically detects target hardware (ESP32, Arduino, Pico) from user prompts.
 
-[Previous changelog content...]
+### 🔄 Phase 4: Feedback Loop
+
+Established continuous improvement cycles.
+
+- **Adaptive Learner**: Created `AdaptiveLearner` to analyze session history for implicit corrections ("actually, use X") and preferences.
+- **Session Analysis**: Added `/analyze` command to scan the current session for learned lessons.
+- **Explicit Teaching**: Added `/teach <rule>` command for manual rule insertion.
+
+### 📈 Metrics & Fine-Tuning
+
+Added tools to measure and cement progress.
+
+- **Learning Metrics**: Created `LearningMetrics` to calculate accuracy trends and correction rates over 30 days.
+- **Fine-Tuning Prep**: Created `ModelFineTuner` to export corrections into JSONL format for training local LLMs (Qwen).
+
+### 🛠 New CLI Commands
+
+- `/learn`: Extract patterns from stored corrections.
+- `/analyze`: Analyze current session for implicit feedback.
+- `/correct <reason>`: Mark previous response as wrong and save correction.
+- `/good`: Mark previous response as correct.
+- `/teach <rule>`: Explicitly teach a coding rule.
+- `/validate`: Run validation checks on the last response.
+- `/rules`: Display currently learned rules.
+- `/metrics`: Show accuracy and improvement stats.
+- `/train`: Export training data for fine-tuning.