A Self-Hosted AI Agent for Finance Operations
Built a company-wide operations platform for 200+ users and deployed a self-hosted Llama agent over the financial data, cutting reconciliation time by 60% and removing 12+ hours of manual reporting a week.
- Role
- Project Lead, Operations Infrastructure & AI
- Organisation
- Pomina Steel Group
- Location
- Ho Chi Minh City, Vietnam
Restructuring a business is hard when nobody trusts the numbers in real time. Pomina's operations ran on spreadsheets and paper approvals, so I led the build of the system underneath it, then put an AI agent on top of the data.
Context
Inventory, manufacturing, sales, and finance each kept their own records. Reconciling them was slow, manual, and always a little out of date, which is exactly what makes capital decisions risky. Management needed to ask a question and get an answer they could act on, not wait a day for a report.
Approach
I led the IT team to build a company-wide operations app for 200+ users across all four functions, and replaced paper approvals with real-time digital notifications.
On top of that data I deployed a self-hosted Llama AI agent. It ran on local infrastructure, so sensitive financial data never left the building. Management could ask about the financials in plain language, auto-fill models, and check metrics on demand.
Self-hosting was the point. For a finance team, an AI assistant is only useful if the data stays in-house, and running Llama locally made that trade-off work.
Result
- Reconciliation time fell by 60%.
- More than 12 hours a week of manual reporting went away.
- Approvals moved from paper to real-time notifications across the business.
What I took from it
The most useful AI in finance is often not a chatbot on a screen. It is the boring infrastructure underneath: clean data plus a well-scoped local agent changed how fast the management team could see and trust their own numbers.