How sukanth-original/what-did-you-get-done works
Project type and purpose
• This is a full-stack, multi-component personal assistant for smart glasses.
• Purpose: capture voice prompts every 15 minutes (or on demand), forward them to an LLM backend, maintain conversational memory, distill memories into structured nodes, and extract reminders.
File structure
Top-level:
• index.ts → real-time UI/server for smart glasses (Node + Bun)
• package.json, bun.lock → JS dependency management
• tsconfig.json → TypeScript configuration
• memory/ → Python micro-services and storage
– llm.py → FastAPI LLM server
– distill.py → memory distillation engine
– reminder.py → reminder extraction utility
– processed_files/ → session tracking persistence
– summaries/, reminders/ → per-user JSONL and text files
README for additional insights
• Four main components are described (UI, LLM, distiller, reminder).
• Core flows: user speaks → index.ts → llm.py → response → display → background jobs (distill, reminder).
• External dependency: Google Gemini model.
• Storage: per-user JSONL files and simple text files.
Final advice
• How user session state flows through processed_files.
• Keep it readable: group related services and separate real-time vs. background.