PlanEat AI
Autonomous nutrition negotiator driven by multi-agent reasoning.
Access is available on request for partners.
The Thesis
Market Context
The global personalized optimization diet market is projected to reach $16B by 2028. However, current solutions (MyFitnessPal, Noom) rely on manual logging, which has a 95% churn rate after 30 days. PlanEat removes logging via Computer Vision and automated negotiation.
Hypothesis
Standard meal planners fail because they are static databases. We hypothesized that an LLM Agent with 'Negotiator Personality' could increase adherence by 300% by adapting to user cravings via active inference.
Technical Challenges
Context Window Saturation
Managing long-term memory of a user's dietary history (30+ days) exceeded standard context windows. We implemented a RAG-based 'Episodic Memory' system to retrieve relevant preferences only when negotiating specific meals.
Non-Deterministic Calorie Counting
LLMs are notoriously bad at math. We solved this by forcing the Agent to call a deterministic 'Nutrition Calculator' tool for all quantitative logic, using the LLM only for qualitative negotiation.
System Design
- 01. User Interface: React Native / Next.js PWA
- 02. Orchestrator: LangGraph State Machine
- 03. Vision Node: Azure Computer Vision + Custom Fine-tuned YOLO
- 04. Memory Store: Pinecone (Vector) + Postgres (Relational)
- 05. Tool Layer: Instacart API, USDA Food Database
Outcomes
Achieved 42% D30 retention. The 'Fridge Vision' feature now processes 10k+ images daily with 98% food recognition accuracy.
Research Roadmap
Core Negotiation Engine & Text-based Planning
Computer Vision Integration & iOS App
Biometric Integration (Glucose Monitors)
Other Experiments
Let's build
something real.
No more slide decks. No more "maybe next quarter".
Let's ship your MVP in weeks.