Affiliated with
I build products from scratch, from first conversation to production.
I'm a software engineer with deep experience in ML, generative AI, and statistics. I've built computer vision pipelines, real-time quoting engines, sandboxed multi-user platforms, and an agentic system to run business operations autonomously.
I have raised venture capital funding from Entrepreneurs First and grown a startup from scratch. My core expertise lies in technical leadership; bridging the communication gap between stakeholders and engineering teams. I am at my best at the intersection of business value and technical complexity.
Candidates solve challenges using AI tools in a live browser sandbox. The HackerRank replacement for the age of vibe coding.
An 8-bit CPU emulator derived entirely from simulated logic gates, later brought into the browser by bundling the Python runtime through WASM.
AI-powered automations for email triage, CRM updates, document processing, and log monitoring. Built for a company on Google Workspace with zero data egress.
A highly specialized code search engine built from scraped GitHub repositories and technical documentation, exposed through an MCP server.
A proof should be an epiphany from the Gods, not a coded message from the Pentagon
No if statements, no control flow — just NAND gates and boolean logic all the way up
Available for software engineering roles and consulting engagements. Amsterdam - San Francisco.
Traditional coding assessments are obsolete because AI writes the code now. VibeArena is a HackerRank/LeetCode replacement for the AI era. The skill that matters is whether someone can effectively use AI tools to solve real problems.
Users get dropped into a browser-based VS Code environment with AI coding tools pre-installed. They solve challenges using AI assistance, and server-side tests verify their solutions. The platform includes progress tracking, leaderboards, and user tiers.
Each challenge spins up a sandboxed Daytona workspace with a full VS Code editor, terminal, and pre-configured AI tools (Claude Code, Cursor). Candidates write and test code in real time while the platform monitors their progress. Server-side test suites validate correctness automatically.
The frontend is built with React and TypeScript, the backend runs on FastAPI with PostgreSQL, and authentication is handled by Clerk. Sandboxed coding environments are provisioned through Daytona. The entire stack is deployed on Railway.
This project is an 8-bit CPU emulator built from the bottom up by deriving the machine's behavior from simulated logic gates rather than hard-coding CPU operations directly. I built it after spending too much time in Turing Complete and wanting to recreate that kind of machine from scratch in Python.
The resulting CPU is already Turing complete with a very small instruction set, and I later turned it into a browser experience by bundling the Python interpreter together with the project through WASM.
The CPU has six general-purpose registers plus a special input/output register, and supports four families of instructions: immediate values, copying, arithmetic / logical operations, and evaluation-based control flow. To make it programmable, I also built a simple assembler with labels that compiles assembly-like code into the underlying binary instructions.
The implementation starts with basic logic gates and builds upward into components such as half adders, full adders, an ALU, registers, and comparison logic. Because the design is derived from the behavior of the gates themselves, changing a single gate changes the behavior of the whole CPU.
The browser version reuses the Python implementation directly by packaging the interpreter and game code with WASM, which made it possible to run the emulator and puzzle interface entirely in the browser.
A suite of internal automation tools built for a company running on Google Workspace. The system handled email triage, CRM updates, document processing, and infrastructure log monitoring — all powered by AI agents that could parse, classify, and act on incoming data autonomously.
Incoming emails were automatically classified using AI filters, then routed to task-specific agents. These agents could parse order details into spreadsheets, draft and send context-aware replies, process job candidate applications, and flag anomalies in system logs for auto-remediation. Rules and tasks were fully configurable through an internal dashboard.
Built as an on-premises solution with full Google Workspace integration (Gmail, Drive, Sheets) via OAuth 2.0 service accounts. All data remained within the client's enterprise environment with zero data egress — no customer data ever left their tenant. The system was designed to meet their data residency and compliance requirements from day one.