I’m a technical generalist building AI at Chess.com. I enjoy learning how things work & tinkering with AI.
Cornell CS, AI minor. Spending my evenings studying management at UChicago Booth.
Currently using
Nov 2025 – Present
Sep 2023 – Sep 2025 · 2 yr
real-world problem → technical solution → business value presentation
The home base for everything your AI makes. Skills, artifacts, and instantly hosted sites live in one workspace you own, at URLs you control. Claude, ChatGPT, and other compatible tools read, write, and publish through MCP or CLI. Switch models without moving your work or losing access to what you have already built, and share context and deliverables with your team ad hoc.
Fine-tuned EmbeddingGemma-300M on MIMIC-III clinical notes using temporal and hierarchical contrastive learning to emphasize clinical meaning over writing style. Built a companion FastAPI copilot for semantic note search, evidence-grounded Q&A, cohort discovery, and patient-trajectory analysis. The project includes two released model checkpoints and a public demo using synthetic records.
My first project & my first with real users (20+ daily actives). Developed a real-time automated parking payment system to reduce costly fines, with estimated throughput of $3,500/week in transaction processing via SMS-triggered workflows, integrating automated scripting and multiprocessing for immediate payment execution.
A personal AI operating system you talk to over iMessage via 
photon.codes, with persistent memory, ad-hoc automations creation (turned into deterministic workflows when possible), and an extensive set of tools via 
composio. Underneath is a file system traversable by agents with CLI & tools, append-only Redis Streams, a model gateway with inference-accelerated OS models, and access controls (clean multi-user support). A planner-executor architecture keeps each agent's context small: the planner delegates to focused executors instead of one agent carrying everything.
Built a disease-agnostic research pipeline on PrimeKG that ranks drug-repurposing candidates and exposes auditable drug–gene–pathway evidence. Compared graph-topology scoring with RotatE knowledge-graph embeddings across four cancer types, examining how the methods surface different candidates and how their rankings can be combined. The pipeline scores 7,957 drugs and produces interpretable evidence paths alongside its rankings.
An email-driven data-science worker that turns dataset attachments into analysis and reports. It profiles data, performs exploratory analysis, optionally runs modeling experiments, and replies in the same email thread with findings and a Markdown report. Supports common tabular formats and ZIP archives, with an always-on worker design and allowlisted senders.
Inspired by the founder-mode approach GitLab founder Sid Sijbrandij takes to his own treatment, documented at osteosarc.com. Built for someone close to me managing complex chronic conditions, this private hub brings medical records into one integrated workspace: a clinical timeline, multi-track treatment protocols, lab and imaging trends, molecular results, and searchable documents. Imported MyChart records and uploaded scans support both human-facing views and an agent-traversable keyspace exposed through CLI and MCP, with conversational record access and configurable research agents.
A personal knowledge layer that lets your agents keep learning from every Cursor and Claude Code session. It keeps your agents.md and rules reflecting how you actually steer, without overfitting to any one session, and lets you pull up old conversations with far higher efficacy when a task needs them.
AI-Powered Government Supplier Procurement For Solving Mission Variability & Equipment Complexity
A real-time hospital dashboard system that optimizes post-operative patient care through AI-driven queue prioritization, automated equipment tracking, and seamless nurse-doctor coordination. Built with Palantir Foundry ontology integration and OpenAI-powered smart workflows.
Guard‑railed chat contextualised to their health data with perplexity-like prompt suggestions for guided continuation.
A precision health platform combining an operational clinician dashboard, a real-time clinical decision support system, and a patient-facing application with a voice AI agent for seamless, real-world data collection, all deeply integrated into foundry's ontology and aip platform
A comprehensive trademark conflict detection and domain availability checking system that combines USPTO trademark data with AI-powered risk assessment and domain availability verification.
A medical chatbot powered by CrewAI that combines real-time RAG with interactive document analysis. The system uses specialized agents to retrieve information from trusted medical sources, Reddit experiences, and uploaded medical documents, while providing NotebookLM-style document summaries and topic-specific discussions.
Implemented 2-Player Chess with OCaml!
Created data pipeline for visualizing and analyzing the captains' Whoop longitudinal wearable data to uncover insights into Cornell Wreslting team's recovery, performance, and sentiment.
my favorite reads and videos of 2026
Reading
Watching
3 Ingredients for Building Reliable Enterprise Agents
★ 3 Key Ideas in AI in 2025
How We Build Effective Agents
Posts
continuous learning and skill development
Course on designing and implementing reliable, scalable cloud infrastructure on Google Cloud Platform using best practices for system architecture, fault tolerance, and operational processes.
Course covering essential cloud computing concepts, Google Cloud Platform services, and foundational skills for deploying and managing applications in the cloud.
Comprehensive course on building, debugging, and optimizing modular GenAI agentic applications using DSPy's module-based programming model with MLflow integration and automated prompt tuning.
Course on implementing AI guardrails to mitigate LLM failure modes like hallucinations and data leaks by adding input/output validation controls to applications like RAG-powered chatbots.
Course on building collaborative multi-agent systems with external tool integration, performance evaluation, and human feedback optimization for automating complex business tasks like project planning and content creation.
Course on designing collaborative AI agent teams using crewAI to automate complex multi-step business processes by assigning specialized roles and breaking down tasks across multiple customized agents.
Course on building and designing various AI agents including document reasoning agents, router agents for Q&A and summarization, and research agents with multi-document handling capabilities and debugging controls.
Course on building and deploying production-ready, low-latency AI voice agents using STT/LLMs/TTS plus WebRTC (via LiveKit), with turn-taking/interrupt handling, latency measurement and optimization, voice customization, and cloud scaling for many concurrent users.
LinkedIn
GitHub
