Paper in. Ledger out. The AI-powered lending platform for Sri Lanka’s microfinance and finance companies.
Fused is adding an AI layer to its ledger. It reads an NIC or a collection sheet from a photograph, takes a collection from a QR card, matches the bank statement on its own and interviews a new client on the officer’s phone. Every result still has to balance, or it is refused.
Identity card, photographed
W. M. R. K. Wijesinghe
Date of birth 02.05.1985
Sex M
No. 42, Temple Road, Kandy
198512304567
Client record
- NIC
- 198512304567, new format
- Name
- W. M. R. K. Wijesinghe
- Date of birth
- 2 May 1985, agrees with the NIC
- Duplicate check
- No other client holds this NIC
Waiting for the officer to confirm
Confirmed. Client 000184234 created, pending activation.
Collection sheet, photographed
Collection sheet
Kandy centre 14, 26 Sep 2026
| Borrower | Due | Collected |
|---|---|---|
| R. Kumari | 2,650.00 | 2,650.00 |
| S. Nadarajah | 2,180.00 | 2,180.00 |
| M. Hameed | 2,000.00 | 2,000.00 |
| K. Perera | 2,450.00 | 2,4?0.00 |
Cash counted 9,280.00
Collection batch
- Row 1
- R. Kumari, 2,650.00
- Row 2
- S. Nadarajah, 2,180.00
- Row 3
- M. Hameed, 2,000.00
- Row 4
- K. Perera, amount unclear
Row 4 needs the officer
Confirmed as 2,450.00. Batch of 4 posted, cash equals lines.
Borrower card at the counter
S. Nadarajah
Loan 000184233
Kandy branch
Counter
- Loan
- 000184233, S. Nadarajah
- Due today
- LKR 2,180.00
- Tendered
- LKR 2,200.00, change 20.00
Teller confirms the cash
Receipt RCT-KDY-2026-004319 printed.
Bank statement, imported
Statement of account
Current account ending 4471, 26 to 27 Sep 2026
| 26 Sep | CHQ DEP 001742 | 45,000.00 |
| 26 Sep | TRF IN HEAD OFFICE | 118,250.00 |
| 27 Sep | SERVICE CHARGE | −350.00 |
| 27 Sep | CASH DEP KANDY | 62,500.00 |
Bank reconciliation
- 45,000.00
- Cheque lodgement, JV-KDY-2026-000088
- 118,250.00
- Head office transfer, JV-KDY-2026-000091
- 62,500.00
- Till deposit, receipts of 26 Sep
- −350.00
- No posting found, left for finance
Waiting for finance to confirm
Confirmed. 3 matches recorded, 1 line for a journal.
On the officer’s phone
Fused field agent
What does the shop earn in a typical week?
Around eighteen thousand. More in April, for Avurudu.
Who else lives on that income?
My husband and our two children.
Do you have any other loans at the moment?
No, this would be the first.
Loan application
- Income
- LKR 18,000.00 a week, more in April
- Household
- Four people, two dependants
- Existing loans
- None declared
Drafted for the officer to review
Reviewed by the officer. Application submitted.
The work the AI takes off the branch.
This is where Fused is going next. Each feature hands its result to a person before anything is written, and each is marked with where it stands.
- Coming
NIC OCR
Fewer typing errors at enrolment.
Photograph a national identity card and the client form fills itself: number, name, date of birth and address, in the old and the new NIC format.
The number is still checked against every other client, and the date of birth against the digits, exactly as it is today.
- Coming
Collection sheet OCR
Forty rows read in the time it takes to photograph them.
Photograph the paper sheet after a centre meeting and every row is read into the batch, ready to key in as one.
A row the model is unsure of is flagged for the officer, and the batch still has to equal the cash counted.
- Coming
QR collection
Shorter queues at the counter.
A borrower’s card carries a QR code. Scan it at the counter or in the field and what is due appears, ready to collect and receipt.
The receipt comes from the branch’s own gap-free series, like every other.
- Coming
Auto reconciliation
A month-end that starts matched.
Bank statement lines are matched to ledger postings on their own: by amount, date and reference, and by the pattern of past matches.
A match the model cannot make with confidence is left for finance, and no match posts anything by itself.
- Coming
Agent-assisted client interview
The same careful interview at every branch.
On the officer’s phone, an AI agent leads the client interview through household, income and existing loans, then drafts the application.
The officer reviews every answer, and the scorecard your institution wrote still decides the grade.
- Coming
Ask the ledger
Answers without waiting for a report to be built.
Ask in plain words, such as what Kandy collected this week against what was due, and get the figures back.
It reads the same curated views your reports run on, and nothing else.
The AI proposes. The ledger still decides.
Nothing on this page is a second way into the books. Whatever a model reads, matches or drafts goes through the same rules as a voucher typed at the counter.
- Nothing writes itself
- Every extraction, match and draft is confirmed by a person before it is posted. The AI shortens the typing; it does not sign.
- The same refusals apply
- An entry from a photographed sheet balances or it is refused, exactly like one typed at the counter. There is no second path into the ledger.
- Every action on the trail
- What the model read, what it proposed and who confirmed it are recorded on the same hash-chained audit trail as everything else.
Built on the platform that runs the branch today.
Ten modules on one ledger, live now: from the loan application to the balance sheet.
The platform in fullAbout the AI layer
Which of these can we use today?
None yet. The ledger, the branch counter, field collections, savings, accounting, reporting and approvals run today. The AI features on this page are coming, and this page says so beside each one.
Does the AI post entries by itself?
No. It reads, matches and drafts, and a person confirms before anything is written. What it writes goes through the same refusals as an entry typed at the counter, and every action is recorded on the audit trail.
What happens when the model is unsure?
It says so. An unclear row on a sheet, a statement line it cannot match or a doubtful answer in an interview is flagged for the officer or for finance rather than guessed.
Continue
See the ledger first, then the AI on top of it.
A walkthrough with the team that builds Fused, on your own products and numbers.