LifeMirror is an AI system that analyzes a user's appearance, media, and online presence to generate socially intelligent perception insights. It breaks down perception across signals such as fashion, face, posture, vibe, strengths, weaknesses, and fix-it suggestions.
This repository contains the backend services, analysis pipeline, media ingestion, and agent orchestration for LifeMirror. Frontend/UI is intentionally out of scope for this phase.
- Media ingestion via presigned URLs
- Vision tools (face, fashion, posture)
- Embedding + similarity search (pgvector)
- Multi-agent perception pipeline (LangGraph)
- Fix-it & improvement suggestions
- Analysis history + notifications
- Auth + consent + privacy controls
- Background processing with Celery
- MinIO object storage
- API: FastAPI
- Orchestration: LangGraph
- Models: Vision + Embeddings + LLMs
- Database: Postgres + pgvector
- Object Storage: MinIO
- Workers: Celery + Redis
src/
├─ api/ # HTTP routes (auth, media, perception, fixit, history)
├─ agents/ # Face, Fashion, Fixit, Orchestrator, etc.
├─ tools/ # YOLO, Face tool, Posture tool, Embed tool
├─ db/ # Models, sessions, migrations
├─ workers/ # Celery tasks for analysis pipeline
├─ services/ # Storage + perception helpers
├─ utils/ # Validation, tracing, mocks, logging
Mid implementation stage.
Completed:
✓ Media ingestion & storage ✓ Vision tools (face, fashion, posture) ✓ Bio / Vibe agent ✓ Embeddings pipeline ✓ Orchestrator (LangGraph) ✓ history ✓ Auth + consent system ✓ Celery background workers
In progress:
- Aggregator + Formatter agent
- Notifications
- Guardrails I/O validation
- DSpy prompt optimization
- Eval suite + tracing
- Deployment manifests (k8s)
- Mobile app integration
