Ahmad Rasoulpour | Salt Lake City, Utah

I build AI products
end to end.

Ahmad Rasoulpour
baba | voice-agent pipeline
Farsi >
agent >

Senior software engineer, 7+ years in immigration tech serving 200,000+ monthly users. On my own time I ship complete products: the voice agent above, a live ML/AI-engineering curriculum platform, a 52K-line mobile game. Design, backend, mobile, billing - the whole stack.

7+
years engineering
3
AI products, built solo
11
languages my products speak
80K+
lines shipped in 2026

Selected work

Baba

In development

An AI assistant for immigrant families - it makes the phone calls, reads the mail, and catches the scams.

A parent writes a request in Farsi, Spanish, or any of 11 languages. Claude turns it into a professional English call script, then a real-time voice pipeline (Deepgram speech-to-text to Claude to ElevenLabs speech) places the call to the insurance company or government office. The family gets the transcript summarized in both languages. Also: document scanning with urgency triage, scam detection, a live two-way translator, and US citizenship test prep.

Under the hood: FastAPI backend with 92 endpoints; dual-principal JWT auth (phone-OTP for parents, OAuth for family); webhook integrations with Stripe, Vapi, RevenueCat, and Twilio; metered call-minute billing; Claude Sonnet/Haiku model tiering plus prompt caching to control inference cost. ~21K lines, solo, across web, iOS/Android, Telegram, and WhatsApp.

ClaudeVapiDeepgramElevenLabsFastAPINext.jsExpo / React NativeStripe11 locales + RTL

Belay

Live

A self-paced platform that retrains software engineers into ML and AI engineers.

An 89-lesson curriculum with runnable Python in the browser, a streaming AI tutor that knows which lesson you're on, a visual skill tree with prerequisite locking, and spaced-repetition review. In production at takebelay.com.

Under the hood: Pyodide/WASM executes numpy and pandas fully client-side; the tutor backend is provider-agnostic (Anthropic SDK or any OpenAI-compatible API, streamed over hand-parsed SSE); FSRS scheduler for review; dual-driver database layer (Neon Postgres in prod, SQLite locally); Stripe subscriptions; Dockerized on Fly.io.

Next.js 16PyodideAnthropic SDKSSE streamingFSRSReactFlowStripeFly.ioNeon

Lucky Summons

Pre-launch

A mobile tower-defense game where luck is the mechanic: summon, merge, defend.

A gacha twist on tower defense for iOS/Android: spend gold on weighted-odds summons, drop units onto board slots, merge three identical units into a higher tier, and hold the line against escalating enemy waves. Five unit classes, four color variants each with a distinct passive ability, plus artifacts, spells, a battle pass, and daily quests.

Under the hood: ~52K lines of C# across 180 scripts and 306 commits in Unity 2022 LTS, built as a team of two. A weighted-probability summoning engine with per-attribute soft floors and board-state success decay; a Supabase cloud-save backend written from scratch over UnityWebRequest (PostgREST upserts, JWT auth, deep-link OAuth, debounced offline-tolerant sync); Unity IAP with subscriptions, rewarded ads with GDPR consent, a multi-currency economy; and 35 custom editor tools that procedurally bake UI prefabs and scenes.

Unity 2022 LTSC#URP 2DSupabaseUnity IAPLevelPlay AdsiOS / Android

Smaller builds

Side quests

Under the hood: GameForge - a generative game-asset pipeline (Llama 3.3 JSON-mode stats, SDXL LoRA sprite sheets, Stable Audio sound effects, packaged Unity-ready with generated .meta files). An algorithmic BTC trading system - walk-forward backtester modeling fees and slippage, exchange-side stop orders that survive downtime, deployed on a VPS under systemd. A multi-chain DeFi tax classifier - reconstructs swaps from raw transfers across five EVM chains, scores classification confidence, exports Form 8949-oriented CSVs.

Groq / Llama 3.3ReplicatePythonccxtweb3.pyFastAPI

Day job

I’ve spent my career in immigration tech: first at SimpleCitizen (early-stage startup), then through its acquisition into Fragomen, the world’s largest immigration law firm. Today I’m a senior engineer there, driving an AI document-intelligence feature - an LLM pipeline that reads a user’s uploaded documents and fills out their immigration application, automating roughly 80% of the form work.

Along the way: led the payments migration to Stripe, shipped an in-platform purchase flow that raised revenue 15% in four months, and led frontend architecture serving 200,000+ monthly active users.