Momed-AI

• 

  Main.py starting picture

Inspiration

Traditional language models suffer from severe context degradation and exponential memory decay in long-horizon interactions due to standard Markovian limitations. We were inspired to solve this by bridging theoretical physics and AI, utilizing Hyperbolic Geometry and Fractional Calculus.

What it does

Momid-AGI is a sovereign artificial intelligence framework that replaces linear token memory with a fractional-order memory architecture governed by Caputo difference equations. It dynamically maps context nodes onto a 2D Riemannian information manifold, maintaining zero signal degradation over massive execution steps while integrating seamlessly with the Gemini 3.5 model and Google Cloud Firestore.

How we built it

• Core Architecture & Math: Programmed in Python utilizing Caputo fractional difference equations, Grünwald-Letnikov expansions, and metric tensor calculations.
• AI & API Gateway: Integrated directly with Gemini 3.5 via google-genai for sovereign agent tasks.
• Cloud Infrastructure: Leveraged Google Cloud Firestore for real-time persistent cross-session state logging.
• Security: Implemented local Fernet-encrypted secure vaults for cryptographic keys.

Challenges we ran into

Balancing the non-monotonic memory feedback weights ($\omega_j$) to prevent state divergence while maintaining an asymptotic efficiency metric ($\eta_k \to 1.0$) required rigorous multi-variable calculus and precision tuning.

Accomplishments that we're proud of

• Successfully executing a 50-step fractional state tracking loop with verifiable mathematical stability.
• Achieving non-blocking, continuous cross-session context persistence via Google Cloud.

What we learned

Deepened our mastery of Riemannian manifolds, fractional calculus operators, and advanced LLM API integration workflows.

What's next for Momid-AGI

Scaling the hyperbolic embedding space for multi-agent autonomous coordination and deploying the sovereign architecture to edge hardware.

Built With

• firestore
• gemini
• git
• google-cloud
• markdown
• python