Take your AI pilot to production, inside your own repo
For founders, CTOs, and product leads at small and mid-sized companies in Europe, Asia, and beyond, whose AI pilot works in a demo and stalls before real users. I am Aris Setiawan, with 13+ years shipping web products. I work inside your codebase, so you end up with code your team can read and extend.
I reply within 24 hours and tell you plainly whether I am the right fit.
Project notes
All my client work is under NDA. These notes are generalized: no client names, sectors, or figures.
AI SEO that researches, writes, and tests itself
The problem
Content teams guess which topics and headlines will work, and they learn slowly because every test takes manual effort.
What I built
An AI system that researches a topic on its own, writes the content, runs A/B tests on the versions, and uses each result to choose better next time.
Indonesian legal AI grounded in expert knowledge
The problem
A general model answers Indonesian legal questions from whatever it remembers. That is not good enough for legal work.
What I built
An AI application trained with input from Indonesian law lecturers and professors, so its answers draw on expert knowledge in addition to the base model’s own.
Full architecture for a high-volume AI product
The problem
An AI product that works on a handful of items falls over when it has to process a thousand at once.
What I built
The whole architecture: software, stack, network, and VPS, tuned to take a thousand records in one go. I also improved the harness around the LLM, which is the code that handles prompts, tools, and retries.
Why AI pilots stall
In an RSM survey of middle-market companies (July 2026, as summarized by DK Studio), data quality and integration were the top blockers among teams with limited pilot success. Both are implementation problems, and they are the work I do.
The data is not ready
The pilot ran on a clean sample. Real inputs are messy, spread across systems, and nobody agreed what a correct answer looks like.
It was never wired into the product
A demo runs on a laptop. Production needs auth, error handling, logging, and a way to switch models without touching every feature.
Nobody owns it after launch
Model updates, cost, and failures land on a team that did not build it and cannot read the code.
What I do
Pilot to production
You have a pilot that works in a demo. I add test cases built from your own inputs, logging, error handling, and spend caps, then hand it over.
See AI integrationAn AI feature inside your product
Search, summaries, an assistant that answers from your data, or smart forms, added next to your current code. No rewrite.
See AI developmentAgents and MCP tools
Agents that call your tools and APIs, where a person approves anything that sends, deletes, or charges.
Read how I scope agentsEngineering retainer
A fixed monthly block for AI feature work, model updates, and fixes after launch.
Cursor rollout for your engineers
Your team bought seats and the PRs got messier. I set up project rules and a review habit on one real repo.
See Cursor mentoringHow I work, in plain answers
These are the questions a careful buyer asks before choosing an AI partner.
- Who does the work?
- I lead every project and you talk to me directly. Trusted partners join when a project needs more hands, and AI agents do much of the build, which is how we ship fast.
- What can I check before I hire you?
- All my client work is under NDA, so I show it as the project notes above, with details generalized. For code you can read, see the open-source projects below and this site, which runs on headless WordPress and Next.js.
- How do we know it works?
- In scoping we agree which data the model may see and what a good answer looks like. Test cases built from your own inputs live in your repo.
- Who owns the code?
- The code, prompts, and test cases live in your repository. Under my terms you own the deliverables made for your project once you have paid in full.
- What happens after I stop working with you?
- You get notes on how it fits together, in code your team can read and extend. A retainer for model updates and fixes is optional.
- What about cost and risk?
- Logging and spend limits go in with the first release, so you can see what each feature costs and cap it. Risky steps wait for a person.
Open source you can read
All my client work is under NDA, so the public proof is code. These are the projects I maintain.
nca (native-cli-ai)
A native Rust CLI that orchestrates AI agents across projects, with persistent sessions, git worktrees, and local-first control.
Spec Kit for Cursor
A spec-driven toolkit for Cursor. The /specify, /plan, and /tasks commands turn an idea into specs, plans, and tasks.
Agent Orchestration
An MCP server for multi-agent work, with shared memory, a task queue, and resource locks.
Chaca Scanner
A desktop web security scanner for developers: OWASP Top 10 checks, API exposure, and CMS detection.
RankMySEO
The open-source version of the rank tracker behind rankmyseo.com.
Good fit and poor fit
Good fit if:
- You have a working product or a pilot and want one AI feature real users touch
- A support bot or internal assistant needs to answer from your own docs or data
- A repetitive step like triage, tagging, or drafting replies should run as an agent with a human check
- Your pilot works in a demo and now has to hold up with real users
Poor fit if:
- You need a model trained from scratch, or ML research
- There is no product or data yet, only a plan to add AI
- You need an on-site team in your office
- Your budget is under $5,000
Questions about AI implementation
How much does AI implementation cost?+
It depends on the use case, the data, and how many systems it touches. Projects start from $5,000. After a short scoping call you get a written plan with milestones and a quote before any code is written.
Can you work inside our existing codebase?+
Yes, that is the usual case. I work in your repo, most often Next.js and TypeScript, and add the AI feature next to your current code. You do not need a rewrite.
Which models do you use?+
OpenAI, Anthropic, or a compatible API, behind one server layer. That way you can switch models later without touching every feature.
Do you work with teams outside Indonesia?+
Yes. I am based in Sidoarjo, Indonesia (UTC+7) and work with startups and teams in Europe, Asia, the US, and the Middle East. I work Monday to Saturday, 9:00 to 17:00 WIB, in English or Indonesian, and we set a weekly check-in time during scoping.
Who will I be working with?+
Me. I lead every project and answer your messages myself. When a project needs more hands, trusted partners join, and I tell you before we start. AI agents do much of the build. Who can access your code and data, including partners and AI tools, is agreed in writing before work starts.
Send me your AI project
Tell me what the pilot does today, where it stalls, and who will use it. I reply with questions or a plain answer on whether I am the right fit.
Perusahaan besar di Indonesia? Lihat implementasi AI untuk perusahaan.