Microsoft track
Compliance and grant administration AI
Seoul, KoreaEst. 202410 peoplePre-Series A · Raising USD 350K
Runs grant administration for 450 SMEs on behalf of Korea's national small-business agency, with no staff assigned. 769 projects, zero clawbacks.
Singapore SMEs claim EDG, PSG and MRA through one portal; from H2 2026 these merge into EDGE. Finance managers and the corporate service providers filing for them lose hours deciding which scheme an activity belongs to. A wrong call after award puts the disbursement at risk.
Read, Model, Judge
Three models in one pipeline: OCR the documents, model the company, judge it against the rule text.
CANVAS · OCR
Reads government forms, receipts and contracts into fields. 0.806, top of five models.
LUCID · Rules
Distribution partner
Corporate service or accounting firm filing EnterpriseSG grants for SME clients. You keep the client, we run the engine.
LUMOS · Singapore · Oct 2026
Pilot customer
One paid pilot on EDGE eligibility screening, H1 2027. We build the Singapore rule pack; you supply cases.
LUMOS · Singapore · Oct 2026
Capital
Pre-Series A co-investor with an ASEAN B2B SaaS or GovTech thesis. USD 150 to 350K, closing Q2 2027.
LUMOS · Singapore · Oct 2026
Pick the companies, we set up the meeting: a call before the week, or a meeting in Singapore.
Register to meet LUMOSSaaS on Azure, or on-premise where data must stay in country. One credential connection builds the company model; the OCR model is trained on 50,000+ government documents. Rule text is loaded as data, so the engine has been ported across four Korean funding regimes. Korean UI, English build underway. Azure Marketplace listing in preparation. PDPA-compatible.
Coming to Singapore
D1 · Mon 26 Oct · 10:15-14:45
IMDA at PIXEL
D2 · Tue 27 Oct · 09:30-18:00
Microsoft programme, then SWITCH
D3 · Wed 28 Oct · 09:30-18:00
SWITCH group meetings
D4 · Thu 29 Oct · 13:00-18:00
Open House
Judges spend and eligibility against law and finance rules. 76.9% win rate over 644 cases.
ANCHOR · Evidence
Cites the clause behind each judgment. General models score 0.000.
