Work

Selected projects.

AI platforms, SaaS products, open-source tools, and developer infrastructure, built end to end, mostly solo.

01

Dealer's Dash

Case study →

AI revenue optimization for auto dealerships

Heavy financial-data platform on AWS. Parses raw and scanned dealership financial statements, pairs a fine-tuned LLM with a deterministic math engine so numbers are always exact, and lets non-technical stakeholders query in plain English. ~90% plan-to-action time reduction.

AWS Python LLM fine-tuning LlamaParse NL query
02

Infollion

Medical-data platform

Data-heavy expert-sourcing system for a B2B expert network. Built scraping and enrichment pipelines that pull and structure large volumes of medical-practitioner data, surfaced through interactive dashboards for client due-diligence.

Scraping / ETL Dashboards Data viz Full-stack
03

MarketifyALL

Case study →

Multi-tenant marketing SaaS · 170 active users

AI content, social publishing, email, SEO, landing pages, and lead CRM, all driven by an AI chat command-center, with per-workspace LLM cost metering and Stripe billing. Grew from empty repo to 170 active users.

Next.js Supabase (RLS/RBAC) OpenAI / Claude Vercel
04

Marketify Design Editor

Open-source Canva alternative

Free, open-source design tool with full canvas engine, 1000+ fonts, stock photos, templates, watermark-free export, plus GPT-4 and Claude for AI-assisted design. Core editor kit published on npm.

React TypeScript FabricJS Supabase
05

email-probe

Case study →

Published npm SDK · zero runtime deps

7-layer email-deliverability pipeline — syntax, DNS/MX, disposable, role, typo-correction, catch-all, and live SMTP checks — with batch and queue distribution. Cut marketing-email bounce rate by ~95%. Full unit-test coverage on every layer.

TypeScript Vitest RabbitMQ npm
06

RankOps

Case study →

Autonomous SEO agent

Connects Google Search Console to Webflow CMS: finds underperforming pages, rewrites metadata with Claude + LangGraph, deploys the change, A/B tests it, and auto-reverts if CTR doesn't improve. An agent that governs its own output.

Claude LangGraph Trigger.dev OpenTelemetry
07

RealStateVideo

Case study →

Live SaaS · Stripe billing

Turns property listings into platform-optimized marketing videos: vision + script generation, viral-hook scoring, TTS, word-level caption alignment, automated FFmpeg rendering, and quality-eval that gates bad renders before delivery.

Astro / React Supabase Trigger.dev FFmpeg
08

AI Voice Call Center

Case study →

Autonomous phone agents

Real-time voice over Twilio Media Streams: Deepgram STT, ElevenLabs TTS, Groq-served LLMs, with barge-in for sub-second turns, Qdrant RAG knowledge bases, post-call LLM scoring as an eval loop, and a one-line embeddable widget.

Next.js PostgreSQL Redis Twilio

Also built / penpot-mcp (MCP server for LLM-driven design edits) · Cortex (RAG voice tutor, Dockerized) · a multimodal vision → Qdrant RAG pipeline · an MCP-backed personal-finance dashboard.