Nickita Khylkouski's Home Page

Student and builder in Redwood City - nickitakhy@gmail.com - resume

Research

Currently I am at Cerebral Valley doing data analysis, software engineering, and security engineering.

Security Research

I have found and reported vulnerabilities across startup, enterprise, and public-sector systems, including critical issues in YC-backed companies, government systems, and critical infrastructure.

Data Enrichment

One thing I am interested in right now is data enrichment. I like taking messy video or image data and turning it into cleaner structure that is actually useful.

For example, if somebody gets out of a car, I want a person, a car, an exit event, and the link between them, not just raw footage.

Recruiting Intelligence

One thing I have been working on is ranking people by what they have actually done instead of what they say about themselves. Resumes are noisy. Competition results, commit history, writing, and actual output usually tell you more, and the hard part is figuring out how to collect those signals and weight them well.

That turned into Metric, which I co-founded with my brother. It pulls signals from competitions, repos, talks, and private or paywalled sources, then ranks candidates with scores and evidence for why they placed where they did. A lot of the work is forcing messy qualitative data into something comparable without flattening the signal too much. We backtested it against real outcomes and got 93% accuracy.

Hackathons

I work at Cerebral Valley, where we host hackathons, product launches, and demo days in San Francisco and beyond. Across different events, we have worked with groups like OpenAI, Anthropic, Google DeepMind, Vercel, Meta, and Y Combinator. I have competed in multiple and won 8, including YC and Replicate.

Background

Outside of that, I have spent a lot of time in security research and penetration testing across enterprise, startup, and public-sector systems.

Some of those findings were severe enough that the estimated downside was in the billions.

Another part of my work is teaching. I am an AI Ambassador and Instructor at Inspirit AI, where I taught AI fundamentals to 60 students and reached 2,500+ people worldwide through education.

Projects

• Metric - recruiting intelligence layer for builders. Scrapes competition leaderboards, gated developer communities, GitHub history, paywalled content, and CTF results. Outputs ranked candidates with scores and evidence. 30K+ people ranked. Try to find the easter eggs on the site.
• SF Graffiti Search - semantic search over 87,790 photos taken by San Francisco city inspectors documenting graffiti violations. Each image was described by Gemini Flash so you can search by content, style, or location. Images sourced by walzr.
• cv-rank (private, email me for more info) - debiased applicant ranking with pairwise Elo-style comparisons. It combines LLM-as-judge scoring, rubric scoring, Swiss pairwise rounds, Bradley-Terry fitting, anonymization, position-bias mitigation, debiasing, and borderline review. Papers: Zheng et al. 2023, Prometheus 2024, Sziklai et al. 2022, Chatbot Arena 2024, Shi et al. 2024, and Tripathi et al. 2025.
• Ranky (private, email me for more info) - human-facing ranking for choosing the best photos. It uses pairwise comparison instead of flat scores, so the final order comes from simpler choices with fewer decisions.
• HumanPaste - paste does not work everywhere. It takes clipboard text and retypes it with realistic keystrokes, which is useful in terminals, remote sessions, and apps that accept typing but not direct paste.
• awake - keeps a Mac awake during long AI agent runs, with safeguards so background work does not die when the machine sleeps.
• ultrathink - drug-discovery search system using AlphaFold, ESMFold, MolGAN, RDKit, and simulated randomness inside an evolutionary search loop. The idea was to grade and improve molecules with many AI tools together so they can cover each other's gaps better than a single model.
• DateGuesser-website - weekday calculation trainer built around Conway's Doomsday algorithm, with repeated practice, feedback, and speed modes.
• BoidDynamics - interactive flocking and complex-systems simulation based on Craig Reynolds's Boids algorithm.

Open Source Contributions

Google/guava and Google/mug: Java utility libraries.
Google/guava: cleanup, test hygiene, thread-leak fixes, and build reliability.
Google/mug: regex edge cases, locale-dependent failures, concurrency tests, and reproducible Maven builds.

Google/dotprompt: prompt tooling across runtimes.
Build and test support, runtime parity fixes, package cleanup, docs fixes, and Python or Bazel workflow improvements.

superset: AI agent desktop and infrastructure work.
Desktop behavior, billing flow, telemetry cleanup, and the ElectricSQL or Cloudflare infrastructure path.

Miscellaneous

I like going to the gym, running, and doing track and field throws.

Lifts / marks: 225 bench, 355 squat, 405 deadlift, 6-minute mile, sub-20 5K.

Lifting account: @nk.liftss

Reaching Me

Email
nickitakhy@gmail.com

GitHub
github.com/nickita-khylkouski

LinkedIn
linkedin.com/in/nickita-khy

Here is my resume.