Hire an AI-native full-stack contractor.
Justin Willhite ships production AI systems by directing AI agents. You get the velocity of a small team for the cost of one operator, plus an architect who has shipped 40+ real systems in the last six months.
What I build for clients.
If it can ship as a self-hosted Python or Flask backend, a static Astro site, a PWA, or an AI agent pipeline, I have done it this year. Below is the menu.
Multi-agent systems that do real work on a schedule. Cross-source data, generate content, push to production, alert on signal.
- OpenRouter, Anthropic, OpenAI, Gemini, Grok via a single gateway
- Cost-gated workers with per-tenant budgets
- Postgres-backed state, systemd timers, full observability
PWA-grade consumer products and internal admin systems. Auth, billing, email, push, the works. No framework lock-in.
- Flask or Node backend, Astro or React frontend
- Stripe billing, Resend email, web push, magic-link auth
- Self-hostable, Docker-shipped, runbook ships with the repo
Multi-property observability for traditional search, answer engines (ChatGPT, Claude, Perplexity), and generative engines (AI Overviews, Copilot). The platform I run for my own portfolio is open to clone.
- GSC, Bing Webmaster, SearXNG, IndexNow, PageSpeed, multi-engine AEO citation testing
- Common Crawl backlink authority, Wikipedia Pageviews, GDELT news mentions
- Wikidata entity monitor, JSON-LD audit, freshness audit, page-level scoring
Bare-metal Linux operations across multiple VPS. Docker where it earns its keep, systemd everywhere else. Production-grade ops without a Kubernetes bill.
- Nginx, Caddy, Cloudflare DNS and WAF, Let's Encrypt
- Postgres on Docker, automated backups, observability via Grafana or Beszel
- Cross-VPS service inventory, port-conflict avoidance, deploy pipelines
Recent work, with outcomes.
Three production systems shipped in the last six months. Each one is real, running, and observable.
Built a multi-tenant observability and opportunity-engine platform for an entire web property portfolio plus client sites. Monitors GSC, Bing Webmaster, 5 LLM citation surfaces, page-level extractability, JSON-LD audit, schema drift, Wikidata entities, news mentions, backlink authority, and crawler logs. Produces a ranked queue of concrete actions per property per day.
- 33 nightly workers, 45+ Postgres tables, 16 dashboard routes
- Multi-engine AEO citation testing across ChatGPT, Claude, Perplexity, Gemini, Grok via OpenRouter
- Free backlink authority via Common Crawl webgraph; Wikipedia Pageviews as topical interest signal
- Auto-drafted content briefs from 6 signal sources, queue moves drafts to publish
Self-hosted quarterly estimated-tax dashboard for a single-member LLC. Ingests Novo bank PDFs, joins to Stripe API, categorizes transactions, runs a custom 2026 federal and California tax engine (existing libraries only go to 2025). Cross-checks SE tax to the dollar across two methods. Seven integrity checks block bad numbers from reaching the dashboard.
- PDF ingestion pipeline, Stripe API read-only, reconciliation to the cent
- Own 2026 tax engine for federal and California, cross-verified
- CA SMLLC Schedule C plus $800 Form 3536 handling
- Review queue for ambiguous income, 7 integrity checks before commit
A daily AI-produced MMA and UFC news podcast plus companion blog. End-to-end: scrape news, deduplicate against a ledger, draft script in voice, TTS via Orus, push to Nextcloud media, generate RSS, deploy site, web approval gate. Runs on a single daily timer. First episode published; site is live; Apple and Spotify feed submissions pending.
- Scrape news, ledger-based dedupe, Gemini Flash temp 0.3 for script
- Web approval gate (Tinyauth-protected) before TTS spend
- Orus TTS, range-capable Nextcloud media, RSS regenerate, Astro site rebuild
- Companion blog post per episode, full transcript included for accessibility plus SEO
The way the engagement runs.
Predictable. Async-first. Shipped output every week of the engagement.
You bring the problem. I bring three or four approaches with rough effort and risk per option. Output is a one-page scope, not a sales pitch.
For non-trivial systems. A two-page brief with the stack, data model, integration points, deploy plan, and weekly milestones. You can hire someone else to build it from this; that is fine.
Demo every Friday with what shipped that week. Live system you can poke. Stack, runbook, and source ship with the deliverable, not after.
Runbook, deploy script, env file template, one-page architecture doc. Two weeks of post-handoff response time included for bug fixes and questions.
Three ways to work together.
Project, hourly, or monthly retainer. Pick what fits your scope and cadence. Net 15 invoicing, ACH preferred. Contact for rates and quotes.
Defined scope, fixed deliverable, milestone-based billing. Best when you know what you want and need it on a date.
Open-ended work, weekly invoicing, ten to thirty hours per week typical. Best when scope is exploratory or the system is evolving.
Reserved capacity, priority response, recurring billing. Best when you have ongoing platform work and want a known hand on the system.
Rate depends on engagement type, scope, and timeline. Send a one-paragraph problem statement and target timeline to heywillhite@gmail.com and you'll get a same-day reply on weekdays with a quote or shortlist of options.
FAQ.
Ready to talk?
One paragraph on the problem and the timeline gets you a same-day reply.
Email Justin