Seeking a Summer 2027 co‑op · Toronto
I make data legible and software real.
I’m Zaki Amin Ahmad, a Data Analytics student and full-stack engineer. I build data pipelines and the products around them, from the schema to the shipped interface.
“To live without risk is to risk without living.”
Selected work
Flagship
The Repo Exchange
A simulated stock market for GitHub repos
A simulated market where GitHub repos trade like stocks. Every price comes from six public numbers on the repo, put through one versioned formula that the Node ledger, the Python worker and the Next.js frontend each implement and test against the same fixtures, so any price can be checked against GitHub by hand. Orders are re-priced on the server, rejected if the price moved more than 1% since the ticket opened, and filled in a single row-locked transaction. You can also open calls: even-money stakes on a repo reaching a star count by a date, settled against GitHub.
Transactions
Orders that can’t double-spend
The server re-prices every order, rejects it if the price moved more than 1%, and fills it in one transaction that locks the account row first. Deadlocks retry on their own.
Concurrency
Settlement that can’t pay out twice
Expired calls are claimed with SKIP LOCKED, so overlapping jobs split the work, and each one settles inside its own savepoint so a bad row can’t stall the rest.
Testing
One pricing formula in three languages, kept in sync
The Node, Python and TypeScript copies run one shared fixture suite, which caught two real mismatches: recency in whole vs fractional days, and half-cent rounding.
Security
The browser never touches the database
Every read and write goes through the ledger, which verifies the Supabase token. Row-level security backs it up, and Turnstile guards sign-up, sign-in and reset.
Performance
Flat database load, however many clients poll
A five-second cache with request coalescing holds Postgres to one listings query per tick, no matter how many tabs are open.
Data pipeline
A GitHub ingester that survives rate limits
Hourly on GitHub Actions, with backoff that honors Retry-After, one request per repo even at 5,000 open PRs, and per-listing commits so a cut-off run keeps its work.
- Next.js
- Node / Express
- Python
- PostgreSQL
- Supabase Auth
- Google OAuth
- Resend
- GitHub Actions
- Vercel
- Render
In progress · Research build, private repo
Batin
Options flow, split by who is behind it
An options order-flow engine built around one question: does big money trading against the crowd say where a stock goes next? A nightly Python ETL, TimescaleDB and FastAPI run on one free ARM server, behind a Next.js terminal. Then I tested the signal: point-in-time backtests, then billions of option trades under rules committed in advance, with a held-out year scored once. Ten analyses found no usable edge, and I said so. An eleventh, a separate, narrow pre-earnings rule on fresh data, passed its locked test thinly: right 50.6% of the time, about even odds it is real. I track it live, with no money traded. Batin carries no predictive claim.
Data engineering
4.25 billion option trades on a free server
Four years of trade-level data, 103,200 ticker-days with zero failures, pulled four API requests at a time with resumable, atomic writes on an ARM VM with 2 cores and 12 GB of RAM.
Market microstructure
Orders rebuilt from raw exchange trades
Trades are marked buys or sells from the quote just before them, and trades on the same contract and side within 10 ms are glued into one order. Rebuilt volume matched official end-of-day totals.
Statistics
A pre-registered test that said no
Two tests had their rules committed before the held-out year was scored, and it was scored once. Following big money was right 49.4% of the time (47.9% to 50.8%) over 11,389 events. Both failed.
Research integrity
A look-ahead leak caught before it counted
Same-day data would have given contracts expiring that day, 53% of SPY’s trades, a delta that is just the day’s outcome. A smoke test caught it, and the fix went in as a dated amendment.
Data integrity
A thin pass, tracked live in a tamper-evident ledger
One narrow rule passed its locked test thinly: right 50.6% of the time, about even odds it is real. Every trading evening, readings are logged before the next open, hash-chained and time-stamped by three independent timestamp authorities.
- Python
- pandas / NumPy
- FastAPI
- TimescaleDB
- PostgreSQL
- Parquet
- Next.js
- Docker Compose
- Linux
- Oracle Cloud (ARM)
Smaller builds
A UI you’re allowed to break
A published npm package that turns any React component into a rigid-body physics simulation, with three ways to break it. This row is wrapped in it.
- Open sourcePublished on npm, with a typed React API
- Computational geometryGlass that fractures along a Voronoi diagram
- RenderingPhysics painted from a live DOM capture
- React
- TypeScript
- matter-js
- d3-delaunay
- html-to-image
- Canvas
- npm
Tower defense where walling in the base doesn’t save you
A tower-defense game in Java and libGDX. Enemies route through your maze with A*, and if you wall the base in, they find the cheapest wall and break through it.
- AlgorithmsA* with a breach mode
- TestingPathfinding checked against brute force
- ArchitectureGame logic written from scratch
- Java
- libGDX
- A* Pathfinding
- JUnit
- OOP
A local AI that cleans up after you
A desktop app that files your documents with an on-device LLM. A regex router handles the obvious cases, so the model only runs on the ambiguous ones.
- Applied LLMsA model that can’t invent a folder
- Systems designThe model only runs when a regex can’t decide
- Data engineeringText out of PDF, Word, PowerPoint and Excel
- Python
- Ollama / Qwen2.5 7B
- CustomTkinter
- Regex + ETL
- JSON Schema
Experience and education
- 2024 to present
Computer Science, Sheridan College
Specializing in data analytics: data pipelines, database management, and software architecture. The trading and analytics systems on this page were built outside of coursework.
- 2022 to 2024
Freelance Web Developer
Designed and shipped custom web applications for clients, including a real estate platform with dynamic listing logic and bespoke React components.
Toolkit
- Languages
- JavaScript
- TypeScript
- Python
- SQL
- Java
- C
- C++
- C#
- Frameworks
- React
- Next.js
- Node / Express
- FastAPI
- .NET
- Data
- PostgreSQL
- TimescaleDB
- Redis
- Supabase
- pandas / NumPy
- Tools
- Git
- Docker
- Vercel
- Railway
- Xcode
- Android Studio