§0. Identification — hero

Self-taught.
Production-shipped.█

I design, build, and run production LLM systems on real servers, end to end, solo. The companies in the callouts are mine. The numbers further on are verifiable.

View the work Email me CV (PDF)

FIG. 1 System diagram — typical production deploy

FIG. 1 — How every one of my products actually runs in production.

§B. Selected work — live products + systems

1. FIG. 2.1  //  DashAway live · v2.0

   DashAway

   A subscription SaaS that rewrites AI-generated text into natural human writing, with real-time detection and click-to-fix suggestions. Full commercial product: accounts, billing, error monitoring.

   frontendNext.js 14 + React + TypeScript
   backendFastAPI · Python
   authSupabase (Google OAuth, JWT)
   billingPaddle subscription
   monitoringGlitchTip / Sentry

   → visit dashaway.io

   

2. FIG. 2.2  //  Promptiv live

   Promptiv

   An inverted flight-search travel engine. Give it a budget and a date window; it returns specific destinations with honest "why this fits, here is the catch" editorial. The budget-first inversion is the moat.

   stackPython · Flask · SQLite · YAML
   apireverse-engineered Google Flights
   pricingAppend-only price-history baseline
   historyEarlier incarnation: prompt-rewriting SaaS (FastAPI + React + Postgres, token metering, pricing tiers) before the travel pivot.

   → visit promptiv.io

   

3. FIG. 2.3  //  ACE (Distill Works) production · 4 clients

   Automated Content Engine

   A multi-client content platform that generates and maintains SEO content across client sites with no per-post human writing: a 13-stage LLM pipeline, 5-dimension quality scoring, and a 3-layer fact-verification system that catches hallucinations before publishing.

   pipeline13 stages · 5-dim quality · 3-layer fact verifier
   isolationPer-client DB, creds, cron — zero shared state
   dashboardsMulti-page client dashboards (Flask + Alpine.js + ApexCharts), tier-gated, per-client auth, admin/operator view, embed mode for the demo
   stackPython · SQLite · Flask · Anthropic API · GSC API · WP REST · BeautifulSoup · Alpine.js · ApexCharts
   deliveryWordPress REST + native static + Slack ops bot

   → see the live demo

   

4. FIG. 2.4  //  Nashville Restaurant Radio production · 426 episodes

   Nashville Restaurant Radio

   A production Python pipeline that pulls new podcast episodes from RSS, transcribes them via API, generates styled show notes, and auto-publishes static HTML pages to the live site. 426 episodes live.

   watchRSS feed (Spotify · Anchor · Creators) — idempotent
   publishPer-episode page + homepage hero + episodes-index + counter, all auto
   transcriptsAPI transcription · structured-data SEO schemas · guest/role enrichment
   infranginx · Let's Encrypt · 301 legacy redirects · GA4 · Resend

   → visit nashvillerestaurantradio.com

   

5. FIG. 2.5  //  LinxSuite → AI Call Center built · deployment-ready

   Customer-comms platform: LinxSuite → AI Call Center

   A two-phase build of customer-communication automation for local service businesses. Phase 1 shipped as an SMS MVP; Phase 2 is the voice-first expansion, deployment-ready but not yet commercially live.

   phase 1LinxSuite (SMS MVP, linxsuite.com): missed-call text-back, reminders, review requests. Next.js + Supabase + Twilio. HMAC-verified webhooks (SHA-256). Deployed via nginx + PM2.
   phase 2AI Call Center (voice-first): white-label AI receptionist — call answer, appointment booking, lead qualification. Twilio Voice + GPT-4 + ElevenLabs TTS. FastAPI + Next.js 14, Postgres + Redis, Docker-composed. Dual-mode (OpenAI + self-hosted Hugging Face).

6. FIG. 2.6  //  Look Closer Today autonomous · ~$0.27/ep

   Look Closer Today

   A fully unattended daily pipeline: it selects a NASA open-data lead, writes vision-guided narration, verifies the narration's spatial claims against the actual imagery with a second model pass, renders animated video, and uploads to YouTube. About twenty-seven cents an episode.

   ingest8 NASA data feeds (APOD · EONET · EPIC · GIBS · DONKI · NeoWs · NOAA)
   narrationAnthropic vision + a second structured-output verification pass
   renderManim animations · ElevenLabs TTS · ffmpeg
   unit cost~$0.27 per published episode

7. FIG. 2.7  //  Reddit Pain Radar data product

   Reddit Pain Radar

   A full-stack market-intelligence platform that scrapes 172+ subreddits, runs GPT-4 pain-point analysis, and scores opportunities for new SaaS ideas. Streamlit dashboard for exploration.

   scrapingChunked Selenium + BeautifulSoup pipeline with anti-detection · 97.5% documented success rate
   analysisGPT-4 extracts pain points, classifies severity, ranks opportunities
   stackPython · Streamlit · SQLAlchemy · pandas · NLTK · scikit-learn · OpenAI GPT-4

§C. More builds — all built and deployed solo

3.1

Demo-Site Lead-Gen Engine

Automated pipeline that generated and deployed 1,058 sites across 538 local businesses: prospect research, geo sweeps, Square payments, sales-rep assignment workflow.

3.2

distill-operator

Production Slack bot to query the ACE platform across clients in plain language. 34 automated tests; systemd deploy.

3.3

NASA Earth Pulse

Real-time NASA event data on an interactive 3D globe. TypeScript monorepo, Next.js, Cesium.js, Fastify.

3.4

etof-metrics

Self-hosted, first-party web analytics (Python + SQLite, /api/track endpoint) so client sites avoid third-party trackers.

3.5

Newsletter platform

Custom Python email newsletter app with subscriber analytics and a spam filter.

3.6

NotebookLM batch generator

Headless Playwright + NotebookLM API to batch-produce audio, video, and slide artifacts from markdown source decks, with retry logic.

3.7

Lead-research pipeline

Apollo.io API + Python + CSV orchestration, rate-limit-aware, to source corporate-travel prospects for the luxury brand.

3.8

MealWise in dev

Next.js 15 + Prisma + Playwright grocery price-comparison app. Kroger API plus a custom ALDI scraper (Puppeteer with anti-detection, ~95% accuracy) for multi-store price optimization.

§D. Operator track — before and alongside the code

§E. Contact — direct lines