Computer Science & Mathematics @ University of Maryland | Aspiring Cloud & Backend Engineer

As a student at the University of Maryland, I've been passionate about research, engineering useful projects, attending hackathons, and working on teams to build ideas into reality. Right now, I'm working on developing agentic systems, engineering full-stack solutions, and collaborating with groups on campus. In the future, I'm looking for MLE/SWE roles, and am currently pursuing internships that align with my interests.

• 🤖 AI Engineer at Testudo Agentic AI - Building autonomous LLM agent systems with self-correcting feedback, tool registries, and RESTful Flask APIs
• 🔬 AI Research - Developed activation steering pipelines for LLM bias reduction, improving BBQ neutrality by 60% (Algoverse Research)
• ☁️ AWS Certification - AWS Certified Cloud Practitioner (January 2026)

Languages

Frameworks & Libraries

Tools & Technologies

Next.js | FastAPI | Computer Vision | Multimodal LLMs | 3DGS

• Won $1K at UMD x IronSite 2026 building spatial intelligence system
• Integrated Multimodal Gemini with vectorized queries for object tracking using 3D Gaussian Splatting data
• Built full-stack application combining Next.js frontend with FastAPI backend for real-time spatial analysis

Next.js | TypeScript | LangChain | LangGraph | AWS Bedrock | Browserbase LIVE AT https://api.rent-swarm.tech

• Won 1st place ($2K) in Marshall Wace track + Best Use of Gemini API at Hack@Brown 2026, competing with 120+ teams
• Architected 5-agent RAG system to analyze Craigslist listings, flag predatory leases, and draft negotiation emails
• Migrated workflow from vanilla Gemini API to LangChain/LangGraph with AWS Bedrock knowledge base for cross-agent state management and PDF document analysis via Retrieval Augmented Generation
• Deployed full-stack app to AWS EC2 with Docker containerization and Caddy reverse-proxy, live in 24 hours

Java | Spring Boot | Docker | Oracle Cloud | Vercel | Prometheus | LangChain4j LIVE AT https://code-runner-eta.vercel.app/

• Cloud-hosted IDE on Oracle Cloud/Vercel handling async Java/Python/C code execution
• Docker-in-Docker sandboxing system with multi-layer security (container isolation, resource limits, network restrictions)
• Async Execution Queue with 10-worker thread pool monitored through Prometheus metrics
• UUID-based polling with ConcurrentHashMap to prevent hangs; integrated LangChain4j agent code tools

Python | PyTorch | TransformerLens | Pickle | MechInterp

• Developed Python pipelines for activation steering and ablation on large language models
• Improved BBQ benchmark neutrality by 60% through targeted intervention on model activations
• Engineered comprehensive logging system with pickle serialization, CSV aggregation, and git versioning across 250+ experiments
• Applied Mechanistic Interpretability techniques using TransformerLens hooks to modify model behavior at inference time

C++ | LabVIEW | Python

• Developed C++ automation framework using Win32 API for async real-time hardware interfacing
• Engineered adaptive binary search algorithm achieving 0.001V precision for electromagnetic calibration
• Reduced experimentation time by 70% and improved accuracy by 30% through automation
• Co-authored published paper in JACS on magnetic field analysis

• LLM Bias Research (Algoverse Research): Developed Python pipelines for activation steering and ablation on LLMs, improving BBQ neutrality by 60%. Engineered logging system with pickle serialization and CSV aggregation across 250+ experiments.
• Toxicity Prediction ML (Beykal Lab, UConn): Designed RNN in Keras for toxicity prediction with <0.05 MAE, consolidating SMILES datasets for end-to-end ML pipeline with dropouts and checkpointing.
• Co-Author: Published paper in JACS on magnetic field analysis (DOI: 10.26434/chemrxiv-2024-45q9k)
• Physical Chemistry Automation (Mani Lab, UConn): Built C++ analysis tools and LabVIEW integrations, reducing experiment time by 70% and improving accuracy by 30%