Full‑stack engineer, B.S. Computer Science from San Francisco State, just off an AI internship at RIA Advisory and open to new‑grad roles. I build secure, performant systems — shipped a Flask + React platform syncing 10k+ records/min at Danieli Corus, and built a SOC triage bot that closes 48% of alerts without escalating them.
I graduated with a B.S. in Computer Science from San Francisco State in May 2026, Dean's List. I cut my teeth at Bharat Electronics building IoT monitoring pipelines, spent summer 2025 at Danieli Corus writing a Flask + React data‑sync platform with OAuth2, RBAC, and end‑to‑end encryption, and at Simulanis built React + GraphQL frontends for AR/VR training reaching 5,000+ users. Most recently I spent summer 2026 as an AI intern at RIA Advisory LLC, building AI solutions for enterprise revenue‑management and billing platforms — a three‑month engagement that wrapped in August 2026.
Outside of internships I keep shipping — pro bono delivered T'chaka, a live restaurant platform for a Haitian restaurant in Oakland; CarbonWise, a personal carbon tracker whose Chrome extension classifies your online activity across 14 sites; a Random Forest SOC triage bot; a Dockerized fake‑news classifier and API; and a local‑first password manager whose own security model I later audited and found wanting — that write‑up is on the site.
Right now I'm deep in Audio Jungle Book — using AI to read animal communication, and auditing whether the pretrained audio models the field relies on actually measure what everyone thinks. Across cats, dogs, pigs and wild bats they mostly don't: individual identity leaks at 63–94% and inflates reported accuracy by 10–24 points. The fix shipped as leakcheck, a pip‑installable audit that fails your CI build if a shuffled split sneaks back in. All of it ran on a laptop for $0.
Selected as a Black Hat USA 2025 Scholar — full scholarship to attend the conference in Las Vegas. Actively looking right now — open to new‑grad software, ML, and security engineering roles, available to start immediately. If you're hiring, or you have a hard problem worth building, my inbox is open.
A curated slice of recent work — ML research, full‑stack platforms, client sites, security tooling, AI agents, SaaS previews, and the odd JavaFX game. Open any card: the case files carry the actual measurements, walkthroughs and demos, not just a description.
Conservation biology now runs forest recordings through large pretrained audio models to decide what animals were doing. I audited whether those models measure what everyone thinks — across cats, dogs, pigs, and wild bats. They mostly don't: the models are answering “who is this?” and being graded as though they answered “what is happening?” Then I shipped the fix as a tool.
A full‑stack restaurant management system built pro bono for a Haitian restaurant in Oakland — live menu, reservation booking, Twilio SMS confirmations, admin dashboard, CI/CD on Railway. Real customers, no fees, no SaaS middleman.
AI‑tool discovery engine for students — keyword search, multi‑filter, community reviews. Full security stack, deployed to AWS, validated with a real usability study.
SOC automation with SIEM ingestion, threat enrichment, and a Random Forest triage classifier on 11 engineered features. Flask dashboard for the metrics.
A Snappr‑style booking platform for a solo photographer — multi‑vertical marketing site, filtered portfolio with lightbox, structured pricing, and a real checkout: Stripe deposits, Supabase persistence behind row‑level security, Cloudinary reference uploads, Resend confirmations, and a gated admin surface.
A lead‑gen research agent: crawls property‑management sites, indexes them for sub‑10ms semantic retrieval, then has Claude judge which are missing tenant portals or online rent payments — ranked leads with evidence and a pitch angle.
A Chrome extension classifies what you do online across 14 sites — streaming, shopping, rides, AI chat — and a model turns each into an emissions estimate you can act on.
Local‑first password manager, and a security audit of my own code: the vault really is encrypted — but there is no master password, and the key sits on disk beside it.
Multiplayer web port of the card game with accounts, a lobby, and a real turn‑state engine — shipped as my SFSU Web App Dev class final.
Research project comparing GPT vs. Claude on resume generation — testing whether ATS‑aware prompts actually beat baseline prompts, with a real rubric.
TF‑IDF + logistic regression behind a Dockerized Flask API. 85–90% on the test set — and an honest account of why that number is the least useful thing about it.
Multi‑level desktop game built in JavaFX with Scene Builder. OO architecture, custom physics, and a leaderboard that actually scrolls.
Next.js preview for local businesses — simulator, conversations, bookings, and a small owner dashboard. Prisma + Postgres, mock LLM / SMS / voice / calendar until real keys.
Portfolio & booking site for an LA mobile videographer — Instagram reels auto‑sync into a phone frame, 3D hero, and a contact form that writes bookings to Supabase with an admin inbox.
Selected to attend Black Hat USA 2025 on a full scholarship — recognition for sustained work in information security, with priority access to advanced training, briefings, and the wider security community. The shortest summary of a year of CTFs, papers read, and security side projects that didn't ship in a single line.
Laid out the way I actually think about it — as one die. Scroll it into view, watch the package open, then pick a block to see what's inside.
What I think in. Eight years of switching between them has made me care more about the shape of a program than the syntax it wears.
The layer people actually touch. Component systems, motion that means something, and 3D when it earns the frame budget — this page runs a Three.js scene above the fold.
The part that has to stay up. REST surfaces, webhooks that reconcile rather than trust, retries and circuit breakers around every third-party call.
Schemas I can defend, constraints at the database rather than in the client, and row-level security where the data is worth protecting.
Frozen encoders, honest evaluation, and models that can say how sure they are. Most recently: auditing what pretrained audio models really measure in animal vocalisations.
Threat modelling before features. Hashing done properly, authorisation checked twice, and encryption that survives someone reading the disk.
Everything between "it works on my laptop" and "it works for everyone". Containers, pipelines, and a deploy that takes minutes instead of an afternoon.
Click a block on the die — or tab into it and press enter. Everything here is something I've actually shipped with.
Live feed · updates as soon as I push a new post
Graduated May 2026, wrapped my last internship in August, and I'm actively looking now — new‑grad software, ML, and security engineering roles, available to start immediately. If you've got an interesting team, an annoying bug, or just want to chat — my inbox is open.
—