On-premises AI and decision systems for a large asset-intensive organisation — where every figure traces to its source row.
Every figure traces to its source row, and filtered queries dropped from 168 ms to 2–5 ms.
Context
A large asset-intensive organisation in Thailand holds a 32,753-line fixed-asset register worth THB 21.03 billion and decades of scanned procurement paperwork.
Challenge
93% of assets had no recorded location, 75% were past their useful life, 65% lacked an accounting asset number. Procurement history existed only on paper. Both systems had to run entirely inside the organisation's own network, with personal data never leaving it.
What we built
- An asset decision system: executive dashboard, Thai full-text register, location-gap pricing and annual physical verification workflow
- A 17-rule data-quality scorecard that flagged 89,266 defects, an accounting reconciliation queue and subtree-scoped role-based access
- A procurement document-AI platform: Thai OCR model plus LLM structuring into a 278-field schema with per-field provenance
- Contractor concentration and price analytics over THB 42.9 bn of contract value and 1,901 unit-price lines, with versioned corrections
- Both systems run entirely on-premises on the organisation's own infrastructure
Outcome
Absence of evidence is never shown as good news; every figure traces to its source row; filtered queries cut from 168 ms to 2–5 ms. 640 pages processed at 100% schema validity.
Architecture
Tell us what your business needs.
Speak directly with our solution architects and get a practical architecture, estimate and delivery plan.