Salesforce MVP Hall of Fame (Shivanath Devinarayanan) · Founded 2022
Loan documents into the loan file
Your credit team stops re-typing regional-language documents. Sarvam Vision reads each page, rules check the arithmetic, and a person approves any field the model is unsure of before it touches the loan file.
Founded 2022 · Founder in the Salesforce MVP Hall of Fame · Dubai HQ · Kochi studio
- 01Models · Claude, Gemini, OpenAI, Bedrock
- 02Agentic systems · Agentforce, CrewAI, ADK, A2A
- 03Agentic memory · Neo4j, Cognee, IndyKite
- 04Automation and voice · n8n, Retell
- 05Deployment and governance · identity, evals, policy, audit
The job
Who it is for, and what Salesforce gets.
The decision
One example, drawn.
Which fields on this page belong on the loan application, and which ones does a person need to check?
A two-page Marathi salary slip, scanned on a phone, with a handwritten employer stamp.
- Employer
- Read, confidence 0.97
- Net monthly pay
- Read, confidence 0.94, matches bank credit
- Pay month
- Read, confidence 0.91
- Employee name
- Read, confidence 0.82, queued: differs from applicant spelling
Illustrative. Written to Salesforce.
- Write to the loan file
- Queue for a reviewer
- Wrong document
- Possible tampering
The flow
One salary slip, into the loan file.
Uploaded
The documents arrive
Salary slips, bank statements, land records. Often a phone scan.
Sarvam reads
Fields, each with a confidence
Sarvam Vision reads the page. Rules check the totals and dates.
Jev checks
Wrong document or tampering
Pages that belong to someone else or look edited are flagged.
Salesforce
Fields on the loan file
Fields under 0.90 go to a reviewer first. Each one links to its source page.
The stack
How we would assemble it.
The honest no: Sarvam has no Arabic, cannot voice Urdu, and runs in India by default. If the job needs any of those, we say so before the build.
Before go-live
What has to be true.
A field under 0.90 confidence goes to a person. Nothing posts to the loan file unreviewed in the first month.
The evidence
What the evidence says
Sarvam's Doc Agents lists bank and credit-card statements, KYC forms, land deeds and handwritten pages among supported documents, in 22 Indian languages plus English.
docs.sarvam.ai/docai/getting-started/overviewTen pages per job on Sarvam's workbench. Long statements need splitting. Check accuracy on your own scans before you trust a threshold.
Who runs it
Who would run this, and what changes.
Example
Credit operations head, Pune housing finance company
Today: Credit officers re-type Marathi salary slips and land records into the loan file. Files wait for the typing, not the decision.
After: Fields land on the loan application with the source page linked. Officers check the unsure fields instead of typing all of them.
Lenders and NBFCsExample
Branch credit manager, Coimbatore MSME lender
Today: Tamil bank statements arrive as phone photos. A total copied wrong is found late, if at all.
After: Rules check the arithmetic before a field posts. In the first month, nothing posts unreviewed.
Sarvam with Salesforce
The brief
Name the call your team makes in the wrong language.
We read the process and write down where Sarvam fits, where it does not, and what Salesforce has to hold before anything goes live.
Sarvam is a trademark of its owner. Mindcat is not affiliated with or endorsed by Sarvam AI.