innerdance — a RAG system you can trust to cite its sources
Retrieval-augmented Q&A over a large, mixed corpus — with measured answer quality.
Germany · remote across DACH, US & UK
I design and ship complete web applications end to end — from the core logic to a genuinely great interface, all the way to production. AI and LLMs are a specialty, brought in when they add real value. One person who owns it all, and measures quality along the way.
The problem
AI ships faster than teams can turn it into something real. A demo impresses; a product has to be reliable, usable, and trustworthy in the messy real world — the confident-but-wrong answer handled gracefully, the long wait made bearable, the error caught before a user sees it. Closing that gap is the whole job — and measuring quality as I go is how I keep it closed.
Proof
Two projects that show the loop end to end — measured quality and the interface around it.
Retrieval-augmented Q&A over a large, mixed corpus — with measured answer quality.
Precisely time-synchronised, analysis-ready physiological data — reliable on mobile.
What I do
Most AI consultants deliver a backend script; most agencies deliver a UI with no substance behind it. I build the entire feature — the AI, the interface, and the production plumbing — done well.
01
The whole application, not a notebook — data model, API, UI, and deployment, with an AI/LLM core when the product calls for it. One person owns it end to end, from first prototype to running in production, so nothing falls between specialists.
02
Four years of React/TypeScript craft most AI engineers don't have. Interfaces that make AI genuinely usable: clear flows, perceived-performance for slow model calls, graceful handling of uncertainty — the layer where users decide whether to trust it.
03
The AI done right: RAG (hybrid retrieval, reranking), agents and tool use, prompt and context engineering, streaming. Real integration of the Claude API and modern LLM tooling — not a thin wrapper.
How it works
An honest call. We figure out what's at stake and where the real risk is. No forms, no jargon.
Usually an eval-system concept on your real data: baseline metrics and an error analysis, so we both know exactly where you stand.
I build the feature end to end — the eval loop and the interface people actually use — and every change is measured, not guessed.
A system that gets better with use instead of quietly rotting. Optional ongoing eval maintenance keeps quality from drifting.
Engagements
Your product, built and shipped end to end — web app or AI feature. From the core logic and API to a polished interface and production. One person, the whole thing.
from €15,000
A focused build: turn a specific idea — a RAG assistant, an agent, a chat feature — into a working, well-designed product. Great for a first project together.
project-based
Need senior React/TypeScript or full-stack hands? Day-rate work — AI feature or not.
day rate on request
Keep building after launch: new features, iteration, and quality kept measured as your product grows.
monthly retainer
Indicative starting points — final scope and price are set together.
“The plus most builders skip: I measure quality as I build — baseline metrics, error analysis, LLM-as-judge — so “it works” isn't a guess. That's how end-to-end delivery stays trustworthy.”
About
I'm Hac Hai. I build complete web applications end to end — the interface, the backend, the API, and the path to production — as one person who owns the whole thing. AI and LLMs are my specialty, brought in when they genuinely add value; plenty of what I build has no AI at all — a solid web app is a solid web app.
Most AI consultants hand over a backend script, and most agencies hand over a UI with nothing behind it. I do both — and I measure quality as I build (baseline metrics, error analysis, LLM-as-judge), so “it works” isn't a guess and it can face real customers.
Background: four years building production React/TypeScript in a large Nx monorepo — component libraries, E2E testing, Auth0/Keycloak — plus deep AI and evaluation work: an open-source RAG system with a full evaluation framework, and a research platform running in production for an NGO health clinic, which I delivered end to end as lead developer.
I'm product-centered: I lead with the user and the outcome, not just the code, with a strong intuition for how something feels to use. I work best with people who bring deep domain expertise — you bring the domain, together we work out what to build and what success looks like, and I take care of the technical side end to end.
What I work with
FAQ
Local
I'm based in Bad Lippspringe, North Rhine-Westphalia — local enough to meet in person around the region, and set up to work remotely with teams across Germany, the wider DACH area, the US, and the UK.
51.783° N · 8.817° E
Contact
If you bring the domain expertise — or you're a seed / Series A team that needs strong hands — I'll handle the technical side end to end, or work with your team to ship a product people love to use.