Applied AI engineering
AI engineering.
For answers you
can stand behind.
I build AI chatbots and search features that answer from your documents and show their sources. If you already have an AI feature, I find where its answers go wrong and give your team a way to measure improvements.
Joel Stevick · Independent engineer & author
20 years shipping software · Frostburg, Maryland
Howard’s Lantern · Saved product example
What happens if I get pulled over and I don't have my license on me?
Maryland law says you must have your license with you and show it to an officer who asks. If you don't have it, you can get a $50 ticket. You can pay it without going to court, and there are no points.
Maryland Transportation §16-112(b–c)
Maryland court fine schedule · §16-112(b)
See how the source checks work →
Selected projects
The work, in production.
All in production. Each one is something I designed and built.
Howard's Lantern: a question box that cites the law
A driving-law exam-prep app whose ask box answers a learner's question in two sentences from their own state's manual or vehicle code, with the section underneath.
- 172,486 law rows across all 51 states, fetched verbatim, dated and hashed
- BM25 and vector search fused by rank, a Haiku rephraser and reranker, then a check that refuses any figure not in a cited law
- Source-verified evaluation cases check answers and citations before changes ship
What you can verify: a model-generated answer is checked before the learner sees it. Unknown citation IDs and figures absent from the cited text trigger rejection or a retry. Inspect the checks, test cases, and evaluation method in the public repository.
Try the product · Inspect the pipeline · Read the evaluation method
PalabraEdu: real-time education SaaS
A five-language multiplayer learning platform on Cloudflare Workers and Durable Objects, with payments, notifications and teacher dashboards.
- Built in six weekends
- Teachers can run live learning sessions across five languages
Quorum
An open-source Claude Code plugin that makes a coding agent write a falsifiable plan, then runs six independent review lenses, a judge, and a CI guard before a pull request opens.
Multi-tenant RAG platform on AWS
Hybrid Pinecone retrieval, Cohere reranking, Bedrock model routing, query decomposition and streamed answers, with the whole stack in AWS CDK.
On-device document intelligence
Commercial PII-redaction and RAG authoring apps for macOS and Windows: DistilBERT on ONNX, searchable PDFs, audit trails, signed offline licensing.
Work with me
A clear scope. A working result.
Fixed prices with a written scope: what is in, what is out, and what you own at the end. Inside the scope the price does not move.
I need a website chatbot
AI chatbot for your website
Give visitors answers from your published documents, with source citations and an “I don’t know” fallback. A working widget, deployed in your account.
Discuss your chatbotScope and payment details
One website widget in one language. One-time import of up to 50 text-based documents and 500,000 extracted words. Verified against 30 agreed questions. Source code, handover guide, and 14 calendar days of fixes for defects against the agreed scope after handover included.
50% to start, 50% at handover. Hosting and model usage are paid by you. Scanned documents, automatic sync, logins, integrations, and ongoing maintenance are quoted separately.
I need to trust an existing AI feature
Grounding and eval review
Find where your AI feature gives unsupported answers. Get a written report, a reusable set of evaluation questions, and fixes ranked by risk.
Request a reviewScope and payment details
One existing RAG or language-model feature reviewed end to end. A starter set of 25–40 source-verified cases and a 60-minute walkthrough. A small evaluation runner is included where the stack and review time allow.
Paid at signing. The $5,000 is credited toward a retrieval build signed within 60 days of review delivery. Implementing fixes is quoted separately.
I need AI answers inside my product
Build a retrieval pipeline
Give your engineering team an API that answers from your documents. Custom retrieval, source checks, and repeatable evaluations, ready for your application to call.
Scope your buildScope and payment details
The base build covers one collection of up to 5,000 documents in one or two agreed formats, a retrieval API, hybrid search, reranking, and checks in code that reject unknown citations and unsupported figures. Includes a 50-case evaluation set, automated tests without model calls, staged fallbacks, and deployment to one environment.
Source code, evaluation results, architecture notes, and an engineer handover included. Payment: 40% to start, 40% at the midpoint demo, 20% at handover.
A chat interface, automatic source refresh, private access controls, additional collections, and integrations are scoped and quoted separately. Hosting, model usage, and ongoing operations are separate.
From the workshop
Built it. Wrote it down.
The methods behind the work, written down so your team can use them too.
Agentic Workflows: Building a RAG Pipeline That Cites Its Sources
Build the Howard's Lantern pipeline by hand, one stage at a time, in TypeScript, and compare each stage with the production code.
Coming soon on Amazon
The 25x Developer
Building real applications with AI agents, quickly, without giving up the testing that keeps them working.
On AmazonThe person behind the work
I’m Joel.
I've built software for twenty years: web, mobile and data platforms, a fleet-inspection app at Zonar that I moved from Angular to React in 60 days, and cargo systems for U.S. Customs and Border Protection. Since 2024 I've built production AI systems as a contractor with Pilot & Pixel, and for my own products. I studied applied mathematics at George Washington University and work from Frostburg, Maryland.
Tell me what you're building
Tell me what you need built or what your AI feature gets wrong. I’ll reply within two working days to confirm fit and arrange a short scoping call.
Or reach me on LinkedIn.