MY-AI - Personal AI Assistant
Personal AI assistant platform with RAG, document knowledge bases, and extensible tool integration, designed to support local models (Ollama) or external APIs (Claude/ChatGPT).
System Architecture
Personal AI assistant platform combining Retrieval Augmented Generation (RAG) with extensible tool integration. Designed to support local models (Ollama) or external APIs (Claude/ChatGPT) with a unified interface.
Documents → Vector DB → RAG Retrieval → LLM Gateway → Tool Orchestration → ResponsePrivacy-focused design with local-first data storage, PII redaction, and bearer token authentication.
Key Features
RAG Knowledge Retrieval
Retrieval-augmented generation system with document ingestion, vector search, and citation-based knowledge synthesis.
Tool Integration Framework
Extensible tool allowlist system with parameter validation and allowlist security, supporting external API integrations.
Provider Flexibility
Supports local models (Ollama) or external APIs (Claude/ChatGPT). Goal is to use local or self-deployed models.
Security & Privacy
Security model with PII redaction, bearer token authentication, and local-first data storage policies.
Technical Details
LLM Gateway
Service for language model interactions with streaming support, timeout management, and model routing capabilities.
Document Processing
Automated document ingestion with provenance tracking, ACL management, and vector database integration for retrieval.
Tool Orchestration
Routing system with JSON schema validation, uncertainty handling, and fallback mechanisms for reliable tool execution.
Impact & Results
Developing a secure, extensible personal AI system that combines knowledge retrieval with tool orchestration
Key Achievements
Designed unified AI provider interface with OpenAI-compatible endpoints supporting dynamic provider/model selection
Implemented full document ingestion pipeline with chunking, Qdrant vector storage, and cited answer retrieval
Created extensible tool framework with registry system, parameter validation, and allowlist security
Built SQLite-based request tracking system with unique request IDs and performance metrics
Developed Typer-based CLI with interactive chat, demos, and tab completion support
Technical Highlights
- • API architecture with clear separation of concerns
- • RAG implementation with citation tracking
- • Flexible provider system (Ollama/external APIs)
- • Privacy-focused design with PII protection
- • Extensible tool framework with registry system
- • SQLite-based request tracking with metrics