Advaita Gupta

01 Serve

Advaita Gupta

I build tools for students and I'm learning to do trading research.

Ninth-grade student. Currently building Pathfind.

See projects GitHub

02 Pass

About

I'm Advaita, a ninth-grade student. I build things to learn how they work.

Most of what I know came from building things that did not work at first. I like building for problems I have seen up close.

I'm most interested in machine learning and the math behind it. Trading is a newer interest, and it led to the trading bot.

Advaita Gupta standing on a path in a blue shirt and black trousers

03 Set

Skills

Languages

  • Python
  • Java

Tools and platforms

  • Git
  • GitHub
  • Supabase
  • Vercel
  • Playwright

Learning

  • Machine learning
  • Data analysis
  • Algorithmic trading

What I can do

  • Start small, test early

    I build the smallest version that works, then add to it and test each change.

  • Check my own work

    When a result looks wrong or too good, I rebuild it from the raw data.

  • Write the rules first

    I decide what counts as a pass before I run the test, so I cannot move the line.

  • Use AI tools on purpose

    I use Claude Code and Codex to write code. I decide what to build and check what comes back.

04 Attack

Projects

Pathfind

Founder and developerIn development

Helps high school students find extracurriculars, internships, and other opportunities.

A catalog of 287 opportunities with search and filters, saved items, Build My Path, an AI assistant, sign-in, and a database with row-level security.

  • Supabase
  • Vercel
  • Playwright
getpathfind.com
  • 287 opportunities in the catalog
  • 109 with a source link and a last-checked date
  • 38 Playwright browser tests

ATL

Open-source contributorPull request open

Agentic Trading Lab, an open-source research project on AI trading agents.

Two Sharpe ratios for the same backtest disagreed. I found the cause, fixed it, added tests, and opened a pull request. It is not merged yet.

Pull request Project site

  • 2.09 / 7.23 One backtest, two different Sharpe ratios
  • 3 new tests: 5-minute data, an overnight gap, a bad timestamp
  • 2 files changed in the pull request

Trading bot

Personal projectIn progress

An Opening Range Breakout (ORB) bot for paper trading, with no real money. It started from a new interest in trading.

So far it has lost money. I tested it, found out why, and wrote it down below.

  • Python
  • Alpaca API
  • QuantConnect/LEAN
  • Git/GitHub
How an opening range breakout works

Case notes

I use AI coding tools, Claude Code and Codex, to write much of the code in these projects. I decide what to build, run the tests, and check the results. What went wrong is below too.

Pathfind

The problem
Programs for high school students are spread over many sites, and it is hard to tell which ones you can join.
What I decided
Real programs stay separate from project ideas the site makes up, so a student never mistakes an idea for a program. Profiles sit behind row-level security, so one student cannot read another's data.
What went wrong
The assistant could not tell whether a deadline had passed. It now gets today's date with every question.
Where it stands
287 listings. 109 have a source link and a last-checked date. 76 have an unknown status. 38 Playwright browser tests run against it. Fixing the unknown ones is next.

ATL

My part
Agentic Trading Lab is someone else's project. I am a contributor, not the founder or the builder.
The problem
The backtest comparison used an hourly Sharpe factor. The runs used 5-minute data. One run showed about 2.09 in one place and 7.23 in another.
What I did
I read the timestamps to find the data interval, skipped overnight gaps, and kept the old hourly factor as a fallback. I added tests for 5-minute data, an overnight gap, and a bad timestamp.
Where it stands
The pull request is open, not merged. My notes on it say I had not yet run the project's full Python tests, so its checks come first. Whether Sharpe on 5-minute data is the right measure here is a separate question.

Trading bot

What I built
A Python bot that ports an Opening Range Breakout strategy to Alpaca paper trading, with free data. It cannot trade real money, by design.
What went wrong
Alpaca rejected some protective stop orders as a possible wash trade. In the first 25 audited paper trades, none won. Fixing the orders was needed, but the research later found the strategy loses after costs.
What I tested
I replayed 2016 to 2023, then 2024 onward, with costs. The average trade lost about 0.07 to 0.09 R. One R is the amount risked on a trade. I wrote the pass rules first. No version passed.
What I learned
Paper fills ignore queues and slippage, so paper results look better than real ones. My own audit had dropped two losing trades, so I rebuilt the numbers from the broker's raw orders. The strategy is not shown to work.

05 Point

Milestones

Done

  • Started Pathfind as its founder and developer.
  • Checked Pathfind's first listings against their own pages.
  • Added Pathfind's AI assistant and Build My Path.
  • Opened a pull request that fixes a Sharpe ratio bug in an open-source trading project.
  • Ran the ORB bot on paper, audited it, and fixed its order bugs.
  • Backtested ORB with costs and wrote down that it did not pass.

Now

  • Building Pathfind. It is in development.
  • Deciding what to try next on the ORB bot.
  • Waiting for review on my Agentic Trading Lab pull request.
  • Strengthening Java and Python.
  • Working on AMC math fundamentals.
  • Learning machine learning, data analysis, and algorithmic trading.

Outside the code

Volleyball.

Competition, repetition, timing, and learning to trust a team. The ball on this page plays one rally, from the serve to the point.

06 Your serve

Contact

Let's build something real.

Open to internships, research opportunities, and meaningful AI/ML collaboration.

GitHub