Hoboken, NJ · USA — Graduate Assistant at Stevens Institute of Technology, previously at Tychee and EquityFx.

I write software for things where being wrong is expensive — trade reconciliation, payment rails, clinical risk scores. Right now I'm on the new-grad/internship interview loop, which is the actual reason AppTracker below exists: I got tired of losing track of my own applications, so I built a tracker with a Kanban pipeline and gamified it hard enough that I keep opening it.

pranavauti2003@gmail.com · linkedin.com/in/pranav-auti

Trade Settlement & Reconciliation Engine — two independent trade feeds (an internal book of record and a counterparty feed) get reconciled; anything that doesn't match becomes a "break," and an ML scorer ranks each break by anomaly severity. Phase 1 of 5, tracked in the repo itself:

• Phase 1 — Postgres schema + Spring Boot matching logic against sample data, unit tests
• Phase 2 — Kafka (KRaft) + Docker Compose, Python feed generator, idempotent consumer
• Phase 3 — REST API for breaks + React dashboard + JWT auth
• Phase 4 — anomaly scoring service, severity surfaced in the UI
• Phase 5 — CI across every service, architecture diagram, seed script, demo walkthrough

Concurrent Resource Booking Platform — the whole point of this one is a single bug: two people click "book" on the same slot in the same millisecond, and a naive read-then-write flow lets both win. Fixed it with row-level pessimistic locking (SELECT ... FOR UPDATE) inside a transaction boundary, then load-tested it — 20 simultaneous booking requests, 1 success, 0 conflicts, every run.

AppTracker — job/internship tracker with a 12-stage Kanban board, a scraper that auto-fills company/role/salary from a pasted Greenhouse, Lever, or Ashby URL with no paid API behind it, and an XP/streak/achievement system that exists purely so I don't ghost my own job search. React 18, TypeScript, Supabase, Vite.

FlashBet Protocol — built in ~24 hours at Monad NYC. Live on-chain YES/NO prediction markets on real sports moments; since there's no live feed to hook into at a hackathon, a scripted Brazil–Argentina match generates real events (penalties, corners) and people bet real testnet MON on them through MetaMask, then claim against contracts actually deployed on Monad testnet. Solidity, Next.js 14, wagmi v2, a Node.js oracle running the match.

AvaxPay — a Stripe-shaped payment processor for Avalanche: deeplink checkout, subscriptions, mock USDC/USDT settlement, contracts live on Fuji testnet. Demo video here. Solidity, Next.js.

FinSight AI — type a ticker, get a real research report back: a year of OHLCV via yfinance, volatility and max-drawdown risk metrics, MA20/50/200 + RSI14, and a rule-based bull/neutral/bear call with reasons attached. No LLM in the loop, no invented numbers. Next.js, FastAPI, pandas/NumPy.

SmartFinance — personal finance tracker, run as an actual Scrum project over 4 sprints for a grad Agile Methods course: budget thresholds with live alerts, savings goals, 6-month spend trends, an admin view across users. React, Node/Express, PostgreSQL.

MaternaSense — clinical decision-support tool for early preeclampsia risk screening, built around evidence-based risk stratification for maternal care planning. React, Python, scikit-learn.

TypeScript Python React Next.js FastAPI Java / Spring Boot PostgreSQL Supabase Solidity Docker Kafka AWS

If you're hiring for SWE / ML / fintech, or want to talk about any of the above — email me.