HOMEPOT AI Integration Roadmap¶

Version: 1.1
Date: November 18, 2025
Status: Planning Phase
Target: 2026
Foundation: Personal AI Companion Architecture

Executive Summary¶

This roadmap outlines the strategic path from the current Complete Website Integration milestone to a fully operational AI-Powered Data Analysis Platform for HOMEPOT Client. The roadmap is divided into 5 major phases spanning approximately 9 months.

Key Decision: We will leverage the proven Personal AI Companion architecture (FastAPI + Ollama + ChromaDB + RAG) as the foundation, adapting it for device monitoring rather than building from scratch. This approach reduces development time from 6-12 months to 5-8 weeks for the AI infrastructure phase, and cuts costs by 75-90%.

Foundation: Personal AI Companion Architecture¶

Overview¶

Instead of building an LLM from scratch (which would require \(100K-\)1M, 100+ GPUs, and 12+ months), we will adapt the proven Personal AI Companion architecture developed by the HOMEPOT team. This architecture has been successfully implemented and tested with:

• FastAPI backend for REST API endpoints
• Ollama for local LLM inference (no third-party dependencies)
• ChromaDB for vector-based memory storage
• SentenceTransformer for embeddings
• RAG (Retrieval-Augmented Generation) for context-aware responses
• Multi-layer memory management (short-term + long-term)

Architecture Components¶

Personal AI Companion (Proven)          HOMEPOT AI Service (Adapted)
─────────────────────────────            ────────────────────────────
app/
├── api.py            # FastAPI          → homepot-ai/api.py
├── llm.py            # Ollama           → homepot-ai/llm.py (same)
├── vector_memory.py  # ChromaDB         → homepot-ai/device_memory.py
├── memory_store.py   # JSON storage     → homepot-ai/event_store.py
├── sentiment.py      # TextBlob         → homepot-ai/anomaly_detection.py
├── persona.py        # Chat modes       → homepot-ai/analysis_modes.py
└── config.yaml       # Configuration    → homepot-ai/config.yaml

Key Adaptations for HOMEPOT¶

Why This Approach?¶

Advantages: - 80% code reuse - Core architecture already built and tested - Local LLM - Ollama runs Llama/Mistral locally (no API costs, full data privacy) - Vector memory - ChromaDB implementation proven for RAG - 5-8 weeks to adapt vs 6-12 months to build from scratch - Fine-tune ready - Can fine-tune Llama 3.2 on HOMEPOT data later - No third-party dependencies - Everything runs on-premises

vs Building from Scratch: - Custom LLM: \(100K-\)1M, 100+ GPUs, 12+ months, massive dataset required - Fine-tuning existing: \(5K-\)20K, single GPU, 2-4 weeks, HOMEPOT data only

Data Security & Storage¶

PostgreSQL (Current HOMEPOT Setup): - Location: 100% local Docker volume (/var/lib/docker/volumes/homepot-client_postgres-data/_data) - Network: Isolated to Docker network, not exposed to internet - Authentication: Password-protected (POSTGRES_PASSWORD) - Persistence: Data survives container restarts and system reboots - Backup: Daily automated backups to local storage

Security Features: 1. At-rest storage - All data stored locally on your infrastructure 2. Network isolation - PostgreSQL only accessible via localhost:5432 3. Authentication - Username/password required for all connections 4. Audit logging - All database changes tracked via audit_logs table 5. Encryption support - pgcrypto extension available for sensitive fields 6. Data retention - 6-month rolling window for AI training, older data archived

AI Training Data Collection: - All training data collected from existing PostgreSQL tables - No external data sources or cloud uploads - Export to local JSONL files for fine-tuning - Complete control over data lifecycle

This roadmap is a living document and will be updated as the project progresses and requirements evolve.