Engineering · Markets · Curious

Tanmay Mangal

I build useful software — financial research tools, full-stack products and applied ML systems.

B.Tech CSE student at VIT Bhopal (2028). I like projects where correctness matters: market data, backtests, event access, model inference, agent evaluation.

View my workDiscuss a project

Available for selected freelance projects, and open to internships and AI/ML or full-stack roles.

Tanmay Mangal — portrait
Selected work

Four things I'd show you first.

Chosen for engineering substance, not stack length. Each README says what works and what doesn't.

MAET

Market analytics & paper-trading terminal · NSE / BSE

Case study →Live →Source on GitHub →
Problem
Retail research tools show numbers without saying where they came from or whether a backtest is honest.
What I built
Screener, charting, a strategy lab (backtests, walk-forward, parameter heatmaps) and a paper-trading engine with OCO brackets and Indian equity charges.
Key decision
Backtest rules are evaluated on bar i−1 and filled at bar i's open, so no look-ahead; paper fills require a fresh live quote. The docs state what is and isn't guaranteed.
Evidence
CI on every push; deployed frontend; engineering docs written from the code.
typescriptreact 19postgresqldrizzlebigquerysse

GatePass

Event admission system · QR passes

Case study →Live →Source on GitHub →
Problem
A ticket must be admitted exactly once, even with flaky door-staff connections and several scanners at the same gate.
What I built
Organizer dashboard, per-event roles, gates and admission rules, scanner devices with camera scanning, attendee pass pages, audit log and CSV exports.
Key decision
Admission is decided inside one database transaction backed by constraints, and the scanner fails closed: no server answer means “not verified”. Retries reuse a request id, so a dropped connection never turns into a false “already used”.
Evidence
Live demo (free tier, data resets on restart). Extensive docs: threat model, test matrix, ADRs. Built on top of CompleteTech-LLC/qr-code-toolkit.
typescriptnodesqlzxing (wasm)docker

Crucible

ONNX inference engine from scratch

Live (browser demo) →Source on GitHub →
Problem
I wanted to understand what actually happens between a .onnx file and a prediction.
What I built
A C++17 engine with its own protobuf parser, graph executor and 20 operators; a FastAPI model-serving service (pybind11), a Rust CLI, and a Rust/WASM interpreter that runs a fraud model in the browser.
Key decision
Unsupported operators and attributes are rejected with an error naming the node, rather than computed wrongly. Output is checked against ONNX Runtime on real models.
Evidence
Honest about limits: single-threaded and several times slower than ONNX Runtime. Browser demo is live; the hosted API is not deployed.
c++17rustwebassemblyfastapipybind11react

Flight Agent Evaluator

Reliability & safety evaluation for tool-using AI agents

Source on GitHub →
Problem
Scoring only an agent's final answer misses the failures that matter: acting after approval was denied, duplicating a payment on retry, claiming success when nothing was committed.
What I built
A deterministic, stateful simulated airline back end plus an evaluator that scores the whole trajectory — tools, arguments, order, side effects and whether the answer is backed by what happened.
Key decision
Everything runs offline and deterministically; the same inputs produce byte-identical reports, and the evaluator's own accuracy is measured against labelled traces.
Evidence
CI on Python 3.11–3.13; benchmark and validation commands reproduce from the repo.
pythonpydanticpytestuv
More work

More projects.

Automated Code Review Tool

Self-hostable PR reviewer: GitHub webhooks, diff parsing, rule-based anti-pattern detection, Next.js dashboard, GitHub Action and VS Code extension. Detection is rule-based by default; the CodeBERT pipeline ships without a trained checkpoint.

View on GitHub →
java 21spring bootpythonfastapinext.jspostgresql

RegimeShift

Offline research lab testing whether regime models improve risk control after costs. Its own finding: the HMM allocation does not reliably beat equal weight. Reproducible CLI and negative controls included.

View on GitHub →
pythonpandashmmstreamlit

Lumint

Cross-modal fraud detection — document, URL and UPI analysis in one pipeline. Built at the Canara Bank SuRaksha Hackathon 2.0.

View on GitHub →
pythonfastapinext.jstypescriptocropencv

FinCalc India

Six financial calculators for Indian users — SIP, EMI, FD, PPF, tax regime comparison.

View on GitHub →View live →
next.jstypescriptpostgresqlprismanextauth

FOSSEE Workshop Booking

Workshop booking platform for the FOSSEE Summer Fellowship @ IIT Bombay, 2026.

View on GitHub →View live →
reactvitetailwindreact-router

AI Image Forensic Screener

Desktop deepfake detector — ML + EXIF + C2PA provenance in one pipeline.

View on GitHub →
pythonpytorchopencvpyqt6c2pa
[ view all repos on github → ]
Services

What I can help build.

  • Web applications

    Responsive sites, business portals, dashboards and full-stack apps.

    See: GatePass, FinCalc India

  • Backend & data systems

    APIs, databases, authentication, data processing and automation.

    See: GatePass, MAET

  • Applied AI & ML

    Practical AI integrations, evaluation tooling and data workflows.

    See: Flight Agent Evaluator, Crucible, Lumint

  • Financial technology

    Dashboards, research tools, screening interfaces and analytical tools.

    See: MAET, RegimeShift, TradeVed internship

Selected projects, one person, direct communication.

Tell me about your project
Professional experience

Where I've worked.

  1. TradeVed Fintech Labs

    Quantitative Research Intern (remote)

    Jun 2026 — Jul 2026

    • Worked on the stock screener's filter logic, data validation and UI workflows
    • Backtested systematic strategies on historical NSE data in Python and Pandas

  2. Dynamic Bubble

    Web Developer Intern (remote)

    May 2026 — Present

    • Built and deployed the company's website (dynamicbubble.agency) from scratch with React, Node.js and MongoDB
    • Keep shipping features, UI improvements and performance work on the live site

  3. Shipd by Datacurve

    AI training data contributor

    May 2026 — Present

    • Designed 10+ coding challenges for Shipd's AI training pipeline (details confidential)

Independent technical contributions

Alongside internships, I take on paid AI-evaluation and data-quality assignments as an independent contributor through Shipd, Outlier, Turing, Meera and DataAnnotation. These are contractor-style tasks, not employment with those platforms, and the client work is confidential, so I don't describe specifics.

Working toward

Internships and roles in AI/ML or full-stack engineering where correctness matters — financial systems, evaluation tooling, products people depend on.

[ show full resume → ]
Markets

Why markets keep showing up in my code.

I started out curious about how markets actually work. That led to trading on Zerodha, then to building my own tools to look at the data, and eventually to the data engineering and ML side of it. Most of my projects still trace back to that.

  • Indian equities and market-data research
  • Company fundamentals and screening (the TradeVed screener is private, so no link)
  • Technical indicators and historical price analysis
  • Backtesting with explicit no-look-ahead rules and trading costs (MAET)
  • Quantitative research and regime analysis (RegimeShift)

RegimeShift's own finding is that the HMM did not reliably beat an equal-weight portfolio. I kept that result in the write-up because reporting it straight matters more than a flattering chart.

This is research and tooling, not investment advice.

About

I'm Tanmay, a B.Tech CSE student at VIT Bhopal graduating in 2028. I build full-stack products, financial-data tools and ML systems, and I care most about the parts that have to be right.

I got into programming because I wanted to understand how markets work. That led to trading on Zerodha, then to building tools I couldn't find, then to machine learning because the questions kept getting harder.

Outside of code: I'm obsessed with food — cooking it, finding it in new cities, thinking about it too much. I travel when I can, and I think staying curious about everything makes you a better engineer.

[ read the resume → ]
Education

Education

VIT Bhopal

B.Tech Computer Science & Engineering · 2024 – 2028

Relevant Coursework

Data Structures & AlgorithmsObject-Oriented ProgrammingOperating SystemsComputer NetworksDatabase SystemsLinear AlgebraProbability & StatisticsMachine Learning

Activities

  • ECell VIT Bhopal — Operations (Nov 2024 – Dec 2025)
  • Techfest IIT Bombay Campus Ambassador (May 2025 – Dec 2025)
Stack

What I work with.

primaryworkingfamiliar

AI / ML
pytorchlangchainmcpgroqopencvocr
Languages
pythontypescriptjavascriptc++sql
Frameworks
next.jsfastapireactflasktailwind
Backend
postgresqlmongodbprismanode.jsexpress
Data
numpypandassqlite
Tools
dockervercelpostmangitgithub
Now

Currently.

Updated as life changes — roughly monthly.

BuildingMAET, GatePass and applied ML / evaluation tooling
Readingmarket microstructure, ML papers, whatever's interesting
Eatingbest food I can find in Bhopal and Agra
Based inVIT Bhopal, Madhya Pradesh
Contact

Have a project in mind?

Tell me what you're building and what you need. Email mangaltanmay7@gmail.com or use the form. I usually reply within a couple of days.

Or message on LinkedIn →

Sent straight to my inbox. I usually reply within a couple of days.

Copied ✓