SERVICES
Three things, done properly. If your problem isn't one of them, I'll tell you and point you somewhere useful.
Agents · RAG · LLMOps
RAG pipelines, LLM integrations and agentic systems trained on your own data, products and customers — written in Python or TypeScript with LangChain, with evaluation and guardrails, so the answers stay correct when the inputs get strange.
Front-end · Back-end · CI/CD
Full-stack build and deployment: React or Next.js on the front, Node.js or Python (FastAPI, Django) behind it, PostgreSQL underneath, shipped through a GitHub Actions CI/CD pipeline to AWS, Vercel or Railway. Empty repository to a product real people log into.
Quoting · Reporting · Documents
The repetitive parts of your operation — quoting, reporting, data entry, document processing — turned into Python pipelines, with pandas and SQL doing the heavy lifting, that run without anyone watching them and tell you when they can't.
…and much more.
STACK
Python and TypeScript, the right model for each job, and a pipeline that checks every change before it ships.
Python
pandas
SQL
PostgreSQL
FastAPI
Django
LangChain
GitHub Actions
Next.js
React
TypeScript
Node.jsLLM apps that hold up in production
AI-native delivery, used with judgement
Python, pandas, SQL
Every change checked before it ships
HOW IT WORKS
The same process whether the project runs two weeks or four months.
01
We work out what you actually need — and whether AI is even the right tool for it. If it isn't, you'll hear that on the call instead of being sold a project.
30 minutes · free
02
Fixed scope, fixed price, a delivery date, and an explicit list of what is not included. In writing, before any work starts.
Within 3 working days
03
You get working software every week, not status reports. Anything heading in the wrong direction gets caught in week two rather than at launch.
Typically 2–12 weeks
04
Deployed, documented, with the repository and the code in your hands. Nothing about the handover depends on keeping me around afterwards.
Included in every project
SELECTED WORK
Every one of these is live and clickable. Open them and try to break them.
Fig. 01—AI agents · Access review · Security
Turns scattered AI-agent permissions into a client-ready access map and least-privilege review — without installing a runtime gateway. Inventories agents, tools and data classes, draws the access graph, flags risky grants, and exports a versioned recertification pack.
Fig. 02—AI BI · MCP · LangGraph
A Model Context Protocol server that gives AI assistants a safe, correct data-analyst capability over business metrics — without improvising raw SQL. Ask questions in plain English; a semantic layer answers them, and nothing the assistant does can damage the data.
Fig. 03—AI agent · Lead gen · Customer support
A self-hosted product: your customers talk to an AI on a public chat page, while you use a private dashboard to review captured leads, upload business knowledge as text or PDFs, and see the meetings the agent booked.
Fig. 04—AI app · LangChain · Next.js
A full-stack app that forecasts used and new vehicle prices over a five-year horizon. Pick a brand and body type; the backend combines historical market data with an LLM analysis pipeline and returns structured yearly projections with confidence scores and depreciation rates.
Fig. 05—Multi-agent · LangChain · Streaming
Paste a URL and several AI personas read the article and argue about it in turn, streaming their replies in real time into draggable agent windows. Built to show multi-agent orchestration and streaming UI in something you can actually play with.
CLIENT FEEDBACK
Reviews from completed engagements, quoted as written.
Lovrobubanic set to the task I requested quickly and had the product completed within a short time. His feedback was prompt and his designs were great.
Great person to work with, did everything and more to make sure the project turned out amazing.
Great outcome, very easy to communicate, fast responses and delivery!
Highly recommended. Very professional and easy to work with. He did exactly what was asked and more, and finished the task 6 days earlier than agreed.
FAQ
The questions that come up on every first call.
Lovro Bubanić — founder, and the only engineer on your project. Neural LB is deliberately a one-person studio: you talk to the person writing the code, there is no account manager in between, and nothing gets quietly handed to a junior after you sign.
THE PERSON BEHIND IT
I'm Lovro Bubanić. I started Neural LB because most AI projects die in the gap between a demo that impresses in a meeting and a system that holds up on a Monday morning with real users and real data.
My background runs across full-stack development and AI engineering, with four games published on Steam from the C# years before that — which is where I learned that shipping is a separate skill from building. Certified through the EITCA Artificial Intelligence Programme at the EITCI Institute in Brussels.
Based in Varaždin, Croatia. I work with founders who need a technical co-pilot to ship fast, and with companies that have an AI idea and no team to execute it.
CONTACT
A couple of sentences is enough to start. You get a reply within 24 hours — and if it isn't a fit, you get told that plainly rather than strung along.
Tap the address to copy it
Calls run on Google Meet, Zoom or Teams. I send the link once we've agreed a time.