Building tools people actually use.
Software Engineer & Product Builder · Applied AI, Backend Systems & Web · Mumbai
featured
The Free University
Un-gated edtech platform and engineering research engine scaled from 0 to 1 with zero marketing spend. Developed a 4-Axis editorial audit rubric to rigorously vet 260+ courses from Google, Harvard, MIT, and IBM across 90+ institution hubs, serving 185 static pages with sub-second response times.
“High-signal technical education was locked behind paywalls and noisy aggregators. Removing paywalls and pre-rendering 185 static pages drove 100,000+ organic learners with zero ad spend.”
mustel
Static analysis tool for AI coding agents. Runs Ruff, Bandit, and pip-audit with 21 custom YAML rules to catch issues before they reach the LLM. Cuts token usage by ~45%.
“AI coding agents waste nearly half their context budget repeatedly fixing basic lint and syntax errors. Intercepting issues locally via MCP before prompt dispatch eliminates that token churn.”
Talanton
Universal token meter and automatic spending limit for AI calls. Enforces hard and soft budget guardrails with sub-2ms local-first latency, zero cloud data exfiltration, and exact pricing calculations for 60+ frontier models.
“Engineers avoid deploying autonomous agents due to fear of runaway billing loops. We bypassed high-latency cloud monitoring proxies by engineering a sub-2ms local-first spending circuit breaker.”
rollit
Rolling window statistics for NumPy arrays, without pulling in pandas. Fast zero-copy calculations with memory-safe array boundary validations.
“Data services shouldn't be forced to pull in a 100MB Pandas dependency just to compute moving averages. Low-level NumPy stride manipulation delivers zero-copy sliding windows instantly.”
Driftwood
Stateless Monte Carlo simulation API for stock price paths, built on Geometric Brownian Motion (GBM). Generates 1,000 price paths in milliseconds with percentile envelopes.
“Financial risk analysis shouldn't require heavyweight, stateful database infrastructure. We turned complex Geometric Brownian Motion simulations into an instant, stateless microservice.”
-d '{"ticker":"AAPL","days":30,"simulations":500}'
lazy footers
Bulk footer stamper for student PDFs and DOCX files. Upload up to 10 files, set the footer text, and download a ZIP. All processing happens in memory.
“Formatting college submission documents shouldn't take hours or risk uploading student research to third-party disks. All batch stamping runs in-memory and destroys on download.”
squeezy
Image compressor and PDF merger built with Flask. Compresses images up to 95% smaller using Pillow, processes everything in memory, and serves downloads via temporary UUID tokens.
“Users shouldn't have to rely on ad-cluttered online file converters that store private images. In-memory Pillow processing compresses images and delivers them via ephemeral UUID tokens.”
experience
- Worked on Project Terminus-2nd-Edition by Snorkel AI, contributing to frontier models as a task reviewer and submitter.
- Evaluated full task submissions and environments (Docker, verifier reports, rubrics) before making Accept / Needs Revision decisions.
- Identified critical issues like hidden answer leakage, non-canonical base images, invalid scoring, and runtime network dependency risks.
- Wrote clear, actionable reviewer feedback on instruction clarity, test alignment, and overall task compliance.
- Built a static analysis tool that runs Ruff, Bandit, and pip-audit for AI coding agents.
- Published on PyPI — 5,000+ downloads.
- Added an MCP server so agents like Claude and Gemini can run analyses directly.
- Built an LLM-based product scraper that reduced manual competitor review time from 4 hours to 1 hour.
- Created a Streamlit + Google Sheets dashboard for tracking intern workloads across the team.
- Evaluated 50+ AI tools for internal product decisions.
- Owned product vision, roadmap, and architecture from 0 to 1, scaling TFU into an edtech product with 260+ editorially vetted courses across 90+ institution hubs (Google, Harvard, MIT, IBM, freeCodeCamp) and 100,000+ lifetime visits with zero marketing spend (featured on Product Hunt).
- Developed TFU's 4-Axis Editorial Audit Rubric to systematically evaluate course Curriculum Depth, Practical Problem Sets, Industry Signal, and 100% Free Access.
- Concepted and launched The Print, a technical research engine publishing 29 deep-dive evaluations across frontier AI models, systems engineering, and cloud infrastructure.
- Created structured career roadmaps (AI Engineer, Backend Developer, Android Developer) and conducted continuous UX iterations optimizing navigation across 90+ provider hubs.
- Architected the platform on Next.js utilizing Static Site Generation across 185 pre-rendered pages to deliver sub-second load times with a 100% un-gated flow (no signups or paywalls).
skills
Languages
Web & Systems
Applied AI & Data
Product & Engineering
about
I'm Ameya Vinayak Kulkarni, a software engineer and product builder based in Mumbai. I work across applied AI, backend systems, and modern web applications with Next.js and TypeScript. I care deeply about product thinking—uncovering real user friction, engineering clean developer experiences, and shipping 0-to-1 products that achieve genuine adoption.
I built mustel, an open-source static analysis tool for AI coding agents with 5,000+ PyPI downloads, and Talanton, a sub-2ms local-first guardrail for AI API spend. I also founded The Free University, scaling it to 100,000+ lifetime visits and 260+ curated courses across 90+ institution hubs.
Currently a third-year IT student at Thakur College of Engineering and Technology (TCET).
currently
- TE IT @ TCET (Semester V)
- Founder, The Free University
- Maintainer, mustel & talanton-py on PyPI
interests
- Running (training for a half marathon)
- Football & jersey collecting
- Open source
