# EdgeMind.AI

- **Event:** [Meta Global AI Developer Hackathon](https://cerebralvalley.ai/e/meta-ai-developer-hackathon)
- **When:** Oct 26, 12:00 PM – Nov 4, 9:00 PM (EST)
- **Where:** Online
- **Team:** [Siddha Shete](https://cerebralvalley.ai/u/siddha_shete26)
- **GitHub:** https://github.com/SID-2324/EdgeMind.ai/tree/main
- **Demo video:** https://youtu.be/_S7eJqmcJWU?si=4HNnyCCEK9CIRE5Y
- **Gallery:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3050

**EdgeMind.ai — Local-First, Hardware-Aware AI Orchestration**

EdgeMind.ai is an intelligent AI orchestration platform designed to solve a growing problem in generative AI: the unnecessary dependence on powerful cloud models, even when a task can be handled efficiently and privately on a user's own hardware.

Traditional AI applications often depend on a single model or fixed execution environment. This creates challenges involving cloud costs, latency, data privacy, internet connectivity, and hardware utilization.

EdgeMind.ai introduces a local-first approach where an orchestration engine evaluates the user's request, privacy requirements, task complexity, system capabilities, and available hardware before determining an appropriate execution strategy.

**Key Features:**

1. **Intelligent Model Routing:** A policy-driven routing system considers task complexity, privacy, resource availability, and execution constraints.

2. **Meta Llama Integration:** Uses Meta Llama 3.2 1B Instruct with OpenVINO optimization to perform real local text and code generation on supported CPUs and GPUs.

3. **Multi-Model Support:** Integrates Meta Llama and Qwen 2.5, demonstrating a modular architecture capable of supporting different AI models.

4. **Hardware-Aware Execution:** Supports CPU/GPU selection and automatic device routing based on model-specific benchmark measurements.

5. **Advanced RAG:** Combines lexical and semantic retrieval with optional reranking to improve contextual information retrieval.

6. **Multi-Stage Orchestration:** Uses LangGraph to coordinate planning, retrieval, routing, model execution, structural verification, and bounded error recovery.

7. **Transparent AI Decisions:** Provides an interactive interface showing results, execution plans, routing traces, and performance information.

**Technical Implementation:**

Built with Python, FastAPI, LangGraph, OpenVINO, Meta Llama, Qwen, and retrieval libraries. The platform has passed 251 automated test cases, including real local model inference and browser integration tests.

The current prototype successfully demonstrates local model execution, intelligent routing, privacy constraints, and workflow orchestration. Cloud inference is part of the proposed architecture but is not yet connected to a live provider.

**Our Vision:**

To make AI more accessible, efficient, and privacy-conscious by allowing applications to intelligently use available computing resources instead of depending exclusively on cloud infrastructure.

EdgeMind.ai is a step toward adaptive AI systems where model selection and execution are determined by the task and environment, not a fixed provider.

## More from Meta Global AI Developer Hackathon

- [Sumora](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3047)
- [Medero](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3048)
- [Clippy](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3049)
- [PairProof](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3051)
- [Christian Friday Douglas](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3052)
- [NearMissLens](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3053)

---

Markdown version of https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3050. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt