Sorcer

Sorcer is a carbon-intelligent AI router to shift LLM workloads to the greenest global grids. Features agentic SDD and real-time routing to ensure high-performance AI never costs the planet.

Comment

Inspiration

We were struck by the paradox of using AI to solve climate change while AI itself consumes more energy than entire nations. We wanted to build a system that treats carbon efficiency not as a trade-off, but as a core technical constraint of the next century.

What it does

Sorcer is a carbon-aware orchestrator that routes and optimizes LLM prompts to the greenest global data centers in real-time. Using Spec-Driven Development (SDD), it identifies non-critical tasks and uses Agentic Deferral to schedule them for "Carbon Valleys" the specific windows when renewable energy peaks on the grid.

How we built it

Our backend is powered by 11Labs, FastAPI, Redis, and ChromaDB, using Vertex AI for model-agnostic routing across Gemini, Claude, and Llama. The frontend is built with Next.js and Framer Motion, and visualizes live carbon data from WattTime and Electricity Maps.

Challenges we ran into

The primary challenge was minimizing the latency added by our carbon-routing engine to ensure the "green" path didn't compromise user experience. We also navigated the complexity of LangGraph state management to ensure our agentic scheduler correctly interpreted technical specs for task prioritization.

Accomplishments that we're proud of

We successfully built a functional end-to-end pipeline that generates "Carbon Receipts" for every prompt, proving the CO2 saved. We are also proud of our cohesive UI, which makes invisible infrastructure data feel tangible and aesthetically striking.

What we learned

We gained a deep understanding of global grid dynamics and the transformative power of Spec-Driven Development. We learned that environmental responsibility can be integrated directly into the software architecture by treating compute as a fluid, temporal resource.

What's next for Sorcer

We plan to implement higher accuracy predictive energy forecasting to optimize Agentic Deferral days in advance. Our ultimate goal is to evolve Sorcer into a developer-first SDK that makes green compute the default standard for the global AI ecosystem.

Built With

  • 11labs
  • and-chromadb
  • and-llama.-the-frontend-is-built-with-next.js-and-framer-motion
  • api
  • chromadb
  • claude
  • fastapi
  • gemini
  • postgresql
  • python
  • react
  • redis
  • using-vertex-ai-for-model-agnostic-routing-across-gemini
Share this project:

Updates