Hi, we're OmniAI.
We're two engineers from Fudan University in Shanghai. We build the software and AI that take repetitive work off your team, and we stay to keep it running after launch.
A lot of good businesses still run on spreadsheets, chat messages and one person who remembers everything. We turn that into simple, sturdy software, and we add AI only where it gives your people their hours back.
Indonesia, steel trading
Every truck that crosses the weighbridge at an Indonesian scrap-steel trader comes with a ticket: supplier, smelter, gross and net weight. That ticket used to be retyped into separate Excel files for invoices, payments and tax, and month-end meant checking them against each other by hand.
Feroxa ERP lets the team enter the ticket once. The invoice, the receipt, the Coretax tax file and the supplier statement all come from that one entry. When something breaks, an AI agent finds the bug, writes a fix and tests it within minutes, and a person approves it before it goes live.
Illustration with sample data
Shanghai, industrial robots
A Shanghai maker of mobile welding robots needed proof of every seam: current, voltage, speed and weave, sent up to a cloud platform. The catch was that the software could look at the robot, but never touch its controls.
The app sits on the robot's industrial PC. It reads the welding data, flags anything doubtful, and serves it to the data box that uploads it. It reconnects and restarts on its own, so it can run around the clock without anyone watching it.
the real screen. 正常 means every value checked out.
China and beyond, family money
Typing in every expense is the part of budgeting people give up on. In Monezi you just say it: takeaway 30, taxi 20. The app hears it, splits it into entries, sorts them into categories and waits for a quick confirm. Snap a long receipt and it reads that too.
Leonardo built it end to end as full-stack engineer: the iPhone, Android and HarmonyOS apps, plus the server, sign-in, payments and the AI behind it, which runs on different models for mainland China and abroad.
Indonesia and Florida, education
At Pelita Harapan Group, Alfin led the development team behind the systems its campuses relied on: online admissions, inventory and facility management, and a Moodle learning platform. Admissions handled more than 4,500 applications a year, and the inventory system kept track of over a billion rupiah in assets.
Before that, at Pensacola Christian College in Florida, he built a course recommendation engine that cut student enrollment errors by 80%, and rebuilt the college's graduation management system.
Fudan University, AI research
AI agents look great in a demo and then fail in strange ways in real use. Alfin's research is about exactly that. He built a pipeline that collected 201 real bug reports from AI agent projects and reproduced 50 of the failures, so fixes can be tested properly. The work was published at NeurIPS 2025, one of the top AI research conferences, with Alfin as co-first author.
It's the same instinct behind the bug-fixing agent in Feroxa ERP. AI only helps a business if it keeps working after the demo.
NeurIPS 202539th Conference on Neural Information Processing Systems
Alfin Wijaya Rahardja*, Junwei Liu*, et al.
* equal contribution
Introduces AgentIssue-Bench, built from real issues in AI agent systems, each reproduced in its own environment so that fixes can be checked automatically.
and plenty more
Upload a floor plan, get a 3D room and furnish it with real products by chatting. The agent remembers your style and budget, and a layout checker stops it from blocking doors.
Leonardo, at Shanghai MaxoTurns photos of metal parts into editable CAD models that help the robot's vision system find each part's exact position. Tested on 1,418 real samples.
Leonardo, Fudan lab projectA prototype that turns a floor plan image into textured 3D rooms, built over one weekend to bring something real to an interview.
LeonardoA POS with stock management for retail, and a restaurant POS for orders, tables and payments.
AlfinA full-stack web app for booking study seats and classrooms, plus an online exam maker for building and running tests.
AlfinTurns a child's conversations with a smart toy into growth reports for parents. Reached the national top 16 of the 2025 Global AI Innovation Competition.
Leonardo, team leadsounds familiar?
If one of these sounds like your business, here is what we would build for it.
We type the same numbers into five different places.
ERP, point of sale, inventory, admissions or bookings. Data goes in once, and invoices, reports and tax files come out of it.
We know AI could help. We just don't know where to start.
We look at how your team works, show where AI saves real hours, what it costs to run, and what to build first. You keep the plan either way.
Half our day goes to reading documents and chasing messages.
Assistants that read documents and receipts, understand voice, answer from your own data, and hand their work to your systems for a person to approve.
Our customers keep asking for an app.
iPhone, Android and HarmonyOS apps with the server, sign-in, payments and admin tools behind them, ready for Chinese and international app stores.
Our machines produce data that nobody ever sees.
Software that safely reads data from robots and machines and puts it where people can use it, plus 3D models from photos and floor plans.
Our AI feature worked in the demo. Real customers broke it.
Test sets, monitoring and fixes for AI features, from the people who published research on how AI agents fail.
no surprises
We talk to the people doing the job and collect the spreadsheets, forms and odd cases that make it hard.
You get something you can click through, using your real data, while changes are still cheap.
Automated tests and release checks on every change, so new features don't break old ones.
Error alerts, user and operations manuals, and fixes when something needs attention.
That's our rule for every AI feature we ship. The agent in Feroxa ERP writes fixes, but a person approves each release. Monezi suggests entries, but you confirm them. The welding software can read the robot, but it can't move it.
the two of us
Co-founder · enterprise systems and AI reliability
Alfin spent seven years in Florida, first studying computer information systems and then building systems for Pensacola Christian College. He went on to lead the development team at Pelita Harapan Group in Indonesia, serving more than 15 campuses. Now at Fudan University, he researches how to make AI agents reliable, and was co-first author of a NeurIPS 2025 paper on it.
seven years in Florida, now building in Shanghai
MS Electronic and Information Engineering, Fudan University, 2024 to 2027, Chinese Government Scholarship
BS Computer Information Systems, Pensacola Christian College, 2013 to 2017
Co-founder · product engineering, AI agents and 3D vision
Leonardo is doing his master's in artificial intelligence at Fudan University, after a bachelor's in AI at Northwestern Polytechnical University, where his thesis on 3D reconstruction was named one of the university's Top 100. He is the sole developer of Feroxa ERP, built the welding robot data system, and was the full-stack engineer on Monezi.
former president of his university's chess club
MEng Artificial Intelligence, Fudan University, 2025 to 2028
BEng Artificial Intelligence, Northwestern Polytechnical University, 2021 to 2025
we speakEnglishBahasa Indonesia中文
so, what eats your team's week?
We'll come back with how we would approach it and what a first version could look like. We keep it in plain language, so you can judge it for yourself.
Reply to the email that brought you here
It comes straight to one of us.