Hi, I'm Suraj

I like tobuild. Software engineer, Forward deployed engineer, GTM engineer, AI engineer.

I take ambiguous problems, turn them into something people actually use, then measure it and keep improving it until it holds up. CS + Data Science at UW–Madison, now building in the Bay Area.

See my work →GitHub ↗

Based in: San Francisco Bay Area
Studied: B.S. CS + Data Science, UW–Madison '26
Experience: 5 roles across startups, media, and research

01 Selected projects

What I've been building lately.

Side projects I've designed and built end to end, with a live demo where there's one to try.

struct-note-suraj-naveens-projects.vercel.app

012026Live demo

StructNote

Walk the job. Talk it through. File the report.

Built for field techs: record a walkthrough on site and get a job report, inspection, or handoff note with the same fields they already file.
Forms first, not a chat window. Empty lines stay empty if you did not say it. No invented parts, names, or measurements.
Presigned S3 upload → SQS → Lambda: Groq Whisper transcribes, Gemini fills the chosen template, and the workspace lets you review, rename, copy, or download markdown.

Next.js
AWS S3 · SQS · Lambda
Supabase
Groq Whisper
Gemini

Try the demo ↗Code ↗

✦ Generated by Sponta AI

9:41●●● ▮

←Today's Challenge⚲

Today's challengeThe Wiki WandererHit “random article” on Wikipedia, then teach a friend one thing you learned.AcademicMedium◉ VerifyRemind me later🔥 5 day streak✓✓✓✓✓SS

022025Launched · 40+ users

SPONTA

A daily dare to go do something new.

An iOS app that gamifies spontaneity for college students with daily challenges, campus events, and streaks.
Built the React Native frontend with a teammate: the onboarding flow, reusable components, a shared styling system, and live input formatting and validation.
Challenges are verified with the camera, synced in real time through Firebase, and generated by Gemini based on category, difficulty, and location.
Launched to 40+ students with a team of four, shipping through reviewed pull requests.

React Native
Expo
Node.js
Firebase
Gemini

View the code ↗

splitsnap-pink.vercel.app

032026Live demo

SplitSnap

Split a dinner bill with friends, before you leave.

Four-step dinner flow: scan or type the receipt, add the table, claim items, then settle. Gemini reads the photo and discards it.
Friends join from a group-chat link with no account. Shared plates split only among who claimed them; unclaimed lines stay out of the math.
Tax and tip follow what each person ordered. Venmo, PayPal, and Cash App open with the amount filled in. SplitSnap never holds the money.

Next.js
TypeScript
Supabase
Gemini
Zod

Try the demo ↗Code ↗

02 Experience

Where I've done the work.

Jan – May 2026
PBS WisconsinSoftware Engineer · Capstone
Staff knowledge lived in 500+ Airtable records. I built a Slack RAG assistant on FastAPI and ChromaDB that answers questions about them in plain English, and owned the part that decides whether those answers are any good: how the records are chunked and embedded.
500+
records staff can search from Slack in plain English
Seconds
to an answer, right inside Slack
Weekly
Agile sprints with stakeholder demos
- Designed the chunking strategy and chose the embedding model, evaluating configurations against real staff queries before each release.
- Gathered requirements directly with staff, then owned the service end to end, from design through deployment.
- Built the Airtable ingestion pipeline with fault-tolerant error handling and schema validation, plus query logging, monitoring, and Tableau dashboards.
- FastAPI
- ChromaDB
- Slack API
- Airtable
- Tableau
Jan – May 2026
Wearable Technologies Inc.ML Consultant · Fall DetectionCase study
A semester-long engagement working directly with the CEO and ML lead. Their production fall classifier was alerting at an implicit 0.50 argmax default that had never been chosen for a care setting, where a missed fall costs far more than a false alarm.
0.50 → 0.40
new alert threshold, adopted by the ML lead
19
thresholds swept across 3 selection policies
+2
real falls caught for every +2 false alarms
- Built a precision-recall sweep with three selection policies, including max F1 and recall-weighted F2, selecting on validation data only.
- Designed and ran a Voice of Customer survey with caregivers at a live care facility on accuracy and wearability.
- Quantified the gap between comfort scores and false-alarm burden; the findings fed product priorities, with a full handoff to the ML team.
- Python
- Precision-recall analysis
- Model evaluation
- Customer research
May – Aug 2025
Dealcycle AIFull-Stack Software Engineer
Shipped two production web apps end to end in Next.js, React, and TypeScript on a microservices architecture.
- Designed the PostgreSQL schema and built the Python and Node.js REST APIs behind the React frontend.
- Owned CI/CD on AWS with Docker through launch and operational support.
- Instrumented 15+ PostHog events, found the onboarding drop-off, and shipped a redesign that improved retention.
- Built a RAG Q&A tool so users could query 20+ client documents.
- Next.js
- TypeScript
- PostgreSQL
- AWS
- Docker
- PostHog
Jan – Nov 2025
MAGIC Lab, UW–MadisonUndergraduate Research Assistant
Research engineering for a university lab running studies across many participants and sessions.
- Built a Python/Pandas pipeline ingesting, cleaning, and analyzing 1K+ data points across 10 studies.
- Wrote PyTorch scripts automating 20+ sessions, cutting setup from 45 minutes to a single command.
- Python
- Pandas
- PyTorch
Jun – Jul 2023
Evenforce TechnologiesSoftware Engineer Intern
Shipped features for Getafix, a garage-management SaaS used by car dealerships, focusing on the speed and reliability of its dashboards and notifications.
4x
faster dashboards, from 8s to under 2s
200+
dealerships using the product
10
person product team, shipping through weekly code reviews
- Profiled and optimized the slow SQL queries behind the notification and dashboard views.
- Improved reliability of the notification system by building real-time REST APIs and debugging its backend SQL.
- REST APIs
- SQL
- Performance

500+PBS Wisconsin records staff can search from Slack

0.40new fall-alert threshold I recommended, adopted by Wear Tech's ML lead

2production web apps shipped end to end at Dealcycle AI

15+PostHog events instrumented to find Dealcycle's onboarding drop-off

45m → 1MAGIC Lab experiment setup, from 45 minutes to one command

4xfaster dashboards at Evenforce, from 8s to under 2s

03 More projects

Side builds and experiments.

Smaller things I've built for fun, at hackathons, or to learn something new. Each one opens its GitHub repo.

04 About

Off the screen, too.

I'm driven to keep raising the bar on what I can build and how much it matters to the people it's for.

That's been the thread through everything I've worked on: a Slack RAG assistant for PBS Wisconsin staff, a fall-detection model audit for a wearables startup, production web apps at Dealcycle, and side projects like StructNote and SplitSnap.

I care about the craft and about the people it reaches. Always happy to grab a coffee and hear what you're working on.

LinkedIn ↗View résumé