scotty ai

Project Concept

No description has been added yet.

Entry

Status: Submitted

Last saved: July 20 at 10:52 AM EDT

View Entry

Team Roster

Message board not available for this team yet.

Ritesh Ojha Team Lead RSVP Approved

Student at New York University

-- Overall Project Leadership & Architecture -- Designed and implemented the complete cloud-native architecture Orchestrated the integration of all components (LLM, agent, frontend, APIs) Managed the entire deployment pipeline and CI/CD process -------------------------------------------------- -- LLM Deployment & GPU Optimization -- Implemented self-hosted Gemma 3 1B deployment using Ollama on Google Cloud Run with NVIDIA L4 GPU Configured GPU settings for cost optimization (single instance, non-zonal redundancy, Europe region) Created custom Docker containers and deployment configurations Optimized model serving parameters for performance and cost efficiency -------------------------------------------------- -- Agentic AI Development -- Built the complete LangGraph agentic workflow using LangChain ecosystem Implemented custom tool-calling logic for open-source models (Gemma 3 1B) Developed multi-tool orchestration system with fallback mechanisms Created the React-style reasoning and action framework -------------------------------------------------- -- Backend Development (FastAPI) -- Designed and implemented the complete FastAPI backend service Integrated with self-hosted LLM via HTTP API calls Implemented comprehensive error handling and logging Created RESTful API endpoints for frontend communication -------------------------------------------------- -------------------------------------------------- -- API Integration & Data Sources -- Google Places API: Integrated for attractions, restaurants, and place search Foursquare API: Added for venue data and recommendations OpenWeatherMap API: Implemented for real-time weather data Exchange Rate API: Added for currency conversion functionality Tavily API: Implemented as fallback search provider Created robust error handling and fallback mechanisms across all APIs -------------------------------------------------- -- Cloud Deployment & DevOps -- Google Cloud Run: Deployed all three services (LLM, backend, frontend) Google Cloud Build: Automated container building and deployment Artifact Registry: Managed Docker container images Cloud Storage: Configured for build artifacts and logs Implemented environment variable management for API keys Created deployment scripts and documentation -------------------------------------------------- -- Security & Best Practices -- Implemented secure API key management via Cloud Run environment variables Used git filter-repo to remove secrets from git history Created comprehensive .gitignore and .dockerignore files Implemented input validation and sanitization Ensured no secrets are committed to version control -------------------------------------------------- -- Performance Optimization -- Optimized GPU utilization for cost efficiency Implemented concurrency settings for optimal scaling Created efficient tool-calling patterns for open-source models Designed fallback mechanisms for API reliability -------------------------------------------------- -- Documentation & Project Management -- Created comprehensive README with deployment instructions Documented architecture decisions and technical implementation Managed the complete project lifecycle from concept to deployment Created clear usage instructions and troubleshooting guides -------------------------------------------------- -- Testing & Quality Assurance -- Conducted end-to-end testing of all components Verified API integrations and error handling Tested deployment pipeline and rollback procedures Ensured cross-browser compatibility and responsive design -------------------------------------------------- -- Innovation & Technical Achievements -- Successfully adapted agentic workflows for open-source models without native tool-calling Created a fully serverless, scalable travel planning solution Implemented real-time data integration with multiple fallback sources Built a cost-effective solution using self-hosted models instead of expensive proprietary APIs -------------------------------------------------- -- Key Technologies & Tools Used -- Google Cloud Platform: Cloud Run, Cloud Build, Artifact Registry, GPU instances Ollama: LLM serving framework for Gemma 3 1B LangGraph/LangChain: Agentic AI framework FastAPI: Backend API framework Streamlit: Frontend web interface Docker: Containerization Git: Version control with security best practices -------------------------------------------------- -- APIs Integrated -- Google Places API Foursquare API OpenWeatherMap API Exchange Rate API Tavily API --------------------------------------------------

Ritesh is an MS Computer Engineering student at NYU Tandon with ~4 years of industry experience spanning ML inference optimization, data engineering, and production AI systems. He has built TensorRT pipelines that cut latency by 35%, deployed vLLM at scale on T4/A100 GPUs, and shipped distributed training jobs with DDP/FSDP. Currently, he's building an autonomous vehicle anomaly detection system using NVIDIA's Cosmos-Reason2-2B and a multi-view camera classification engine. A hackathon regular, he won the Qualcomm Edge AI and Visa Challenge tracks at HackNYU and most recently built ClawFin at the AI Tinkerers NYC ClawHack. He's interested in agentic AI, GPU systems, and anything that ships.

Agentic AI infrastructure, multi-agent orchestration, and runtime enforcement for autonomous agent systems. Interested in spec-driven development for AI agents, scaling GPU inference at the edge, and swarm detection in production environments. Looking to go deeper on Rust for high-performance AI platforms, ontology-backed agent architectures, and zero-trust security patterns for multi-agent workflows. Also exploring serverless and containerized approaches to deploying agentic AI SDKs.

Building an autonomous vehicle semantic anomaly detection system using NVIDIA Cosmos-Reason2-2B with a multi-view camera pipeline and 8-layer classification engine. Developing ClawFin, a multi-agent deal negotiation platform using OpenClaw, XMTP, and OpenRouter. Collaborating on LLM-based homework feedback analysis research at NYU Tandon. Experimenting with zero-cost autonomous pipelines using LangGraph, MCP, and Google Cloud Run with GPU inference for agentic AI workflows.

Eunice Lee RSVP Approved

Student at Carnegie Mellon University

-- Frontend Development (Streamlit) -- Built the complete Streamlit web interface with modern UI/UX Implemented query templates, real-time status indicators, and download functionality Created a responsive design with beautiful styling and user guidance Integrated with the backend API for a seamless user experience

Eunice Lee is a frontend engineer and undergraduate student at Carnegie Mellon University, pursuing a Bachelor of Science in Information Systems with an expected graduation in 2026.

frontend development, user experience design, product development, research, JavaScript, Python, React Native, UI/UX improvements, feature shipping, collaborative coding projects