💡 Inspiration

Modern software delivery pipelines generate hundreds of static alerts daily with high false-positive rates, while manual code audits cannot keep up with CI/CD release cadences. Most security tools simply throw alerts at engineers without remediation steps.

SynapseSentry AI transforms code security from passive warning banners into active, self-healing remediation. Powered by NVIDIA Nemotron-70B and Nebius Token Factory, the agent reasons over abstract syntax trees (ASTs), references live threat bulletins via Tavily AI, and autonomously synthesizes verified, production-ready Git unified diff patches in a single click.

🛠️ How We Built It

Model Reasoning Core: Leveraged nvidia/llama-3.1-nemotron-70b-instruct for deep Chain-of-Thought (CoT) AST decomposition and semantic vulnerability analysis.
Inference Infrastructure: Architected natively for Nebius Token Factory using OpenAI-compatible REST endpoints, with seamless wire-compatibility to NVIDIA NIM.
Live Threat Intelligence: Integrated Tavily Real-Time Search API to fetch live CVE bulletins from NIST NVD and MITRE databases for detected vulnerability sinks.
Full-Stack Dashboard: Built with FastAPI, Python 3.13, and modern CSS glassmorphism, featuring a dynamic Security Health Index dial, real-time threat feed, and 1-click safe code injection.
Mobile-First Accessibility: Engineered to run smoothly on constrained mobile developer hardware (Android via Termux and Acode editor) without heavyweight compilation toolchains.

⚡ Challenges We Ran Into

Generating strictly deterministic, syntactically valid Git unified diff patches (git apply compliant) from deep reasoning models can often suffer from whitespace drift. We solved this by implementing strict JSON output schemas with Chain-of-Thought validation bounding.

🏆 Accomplishments That We're Proud Of

Designing a complete, interactive cyber-command center featuring a 1-click self-healing code remediation loop that automatically upgrades vulnerable code to a 100/100 Safe posture.
Seamlessly marrying NVIDIA open models with live global OSINT threat intelligence through Tavily AI.
Developing and deploying the entire project from a mobile smartphone environment.

🔮 What's Next for SynapseSentry AI

GitHub Actions & GitLab CI native runner integration for automated PR scanning and auto-patch approvals.
Jetson Orin Nano edge deployment for IoT firmware security analysis.

Built With

css3
cybersecurity
fastapi
git
html5
nebius-token-factory
nvidia-nemotron
python
tavily-api
uvicorn

Submitted to

Nebius x NVIDIA Global AI Hackathon

Created by

As the solo architect and developer of SynapseSentry AI, I conceptualized, designed, and built the entire project end-to-end:

1. System Architecture & Model Pipeline: Designed the core reasoning engine utilizing NVIDIA Nemotron-70B for AST decomposition, security vulnerability classification (CWE-89, CWE-78, CWE-798), and atomic Git unified diff synthesis.
2. Cloud & Threat Intelligence Integration: Configured the Nebius Token Factory wire-compatible API client and integrated the Tavily Real-Time Search API to dynamically pull live CVE threat bulletins from NIST NVD and MITRE databases.
3. Interactive Cyber Dashboard: Built a full-stack web application using FastAPI, Python 3.13, and modern CSS/HTML with a real-time Security Health Index dial, live threat feed, and a 1-click self-healing code remediation workflow.
4. Mobile-First Engineering: Optimized the entire codebase to run smoothly in constrained developer environments, developing and testing directly on an Android smartphone via Termux and Acode editor.
5. Project Deliverables: Produced all project documentation, open-source MIT licensing, demonstration scripts, and video materials.

Dawar Wani

Updates

Dawar Wani started this project — Oct 27, 2026 07:22 AM EDT

Leave feedback in the comments!

Log in or sign up for Devpost to join the conversation.