Back end
- Go Fiber
- FastAPI
- MySQL
- PostgreSQL
- Redis
Two years building a digital lending platform used by banks, finance companies and their partner agents, from React screens to Go back ends.
Go · React · TypeScript · MySQL
Who I am
I’mafull-stacksoftwareengineerwithovertwoyearsofexperiencebuildingadigitallendingplatform.Banks,financecompaniesandtheirpartneragentsuseittotakeinloanapplications,checkcustomerdocuments,approveloansandmanageeachloanfromstarttofinish.Iworkonbothsides:thescreenspeopleuseeveryday,andthebackendthatstoresdata,appliesbusinessrulesandproducesreports.Ialsohandleclientsdirectlyandworkcloselywithfinanceandoperationsteamstoturntheirneedsintoworkingsoftware.
Years in FinTech
Reviewed code changes
Lines of code written
Business areas covered
What I do, in one line
Hover to x-ray →
I build the software behind every loan.
At a glance
4 core strengths
Tools of the trade
Keep scrolling →
Inforvio Technologies
Full-stack work on a digital lending platform used by banks, finance companies and partner agents: loan applications, document checks, approvals, accounting and the full loan lifecycle. I also handle clients directly.
itboomi innovations
Built features with MongoDB and applied SEO best practices to web projects.
The next chapter
I’m looking for a FinTech or product team where I can keep shipping lending and finance software end to end. Let’s talk.
Get in touch →Since September 2024 I’ve worked on a lending platform that banks, finance companies and their partner agents use to run loans end to end.
20+ business areas · Lender API integrations · Go + React
Key achievements
06 highlights
01 · Recent · Two-wheeler loans
Connected the platform to a lending partner’s APIs for two-wheeler loans, covering every step from creating the application to vehicle delivery. I also integrated the IAAS API.
8-step loan journey, end to end
02 · AI / OCR integration
Customer documents such as Aadhaar, PAN, GST certificates, cheques, driving licences and utility bills are read automatically, and unclear or wrong documents are flagged. This removes manual typing and speeds up loan processing.
15+ document types
03 · Credit appraisal · PDF
The report loan officers use to decide on a business loan. It pulls together tax returns, GST data, credit score, financial health, risk rating and the loan offers available.
One ready-to-print PDF
04 · Workflow & approvals
A system for after-loan requests such as a loan closure quote or a change of details, with rules for who can approve what, required documents and a full history for each request.
Raised → approved → completed
05 · Finance automation
Monthly commission for partner agents is calculated automatically, with PDF invoices and GST records generated each month.
Replaced manual spreadsheets
06 · Config-driven product
Business teams change loan application steps and forms through settings, so new products and changes go live without waiting for a release.
No developer needed
01 / 06 · Lender API integration
I’ve built features at every stage of a loan, from the first application screen to recovery, so I understand how the pieces connect.
Stage 01
Stage 02
Stage 03
Stage 04
Stage 05
Stage 06
The full list
Everything I’ve worked on across 20+ business areas of the lending platform, grouped the way the business uses it.
Main project · Working MVP · Started Nov 2026
Python · FastAPI · GPU OCR
OCR API for Indian KYC and loan documents
An API that reads 14 kinds of Indian KYC and loan documents (photos, scans or PDFs) and returns checked, structured JSON in about 0.2 seconds, with a clear flag when a person should look.
Permanent account number · sample
ABCPR1234F
Aadhaar · masked by default
XXXX XXXX 4821
Sample bank · cheque
IFSC ABCD0123456
₹ 1,25,000.00
⑈004512⑈ 600002000⑆ 11
GST registration · sample
GSTIN 33ABCPR1234F1Z5
Bank statement · Apr–Sep
Sanction letter
PDFLoan sanctioned
Amount
₹5,00,000
EMI
₹16,251.22
Rate
10.5% p.a.
Tenure
36 months
The problem
Lenders still type PAN, Aadhaar, bank statement and salary slip details by hand, or pay per document for OCR that reads text but doesn’t check it. A wrong digit in a PAN or a doctored bank statement goes straight into the loan file.
What I built
A production-style API: open-source OCR models on the GPU, a validator for every number (Verhoeff, GSTIN, MRZ check digits, IFSC), bank statement analysis, API keys with quotas, background jobs with signed webhooks, and monitoring. Built and tested end to end on a laptop at zero cost.
$ POST /v1/documents:extract pan.jpg
{
"document_type": "pan",
"fields": {
"pan": { "value": "ABCPK1234F", "confidence": 1.0 },
"name": { "value": "RAVI KUMAR", "confidence": 0.992 },
"dob": { "value": "1990-04-12", "confidence": 1.0 }
},
"checks": {
"pan_format": true,
"name_matches_pan": true
},
"needs_review": false,
"processing_ms": 148
}Real API responses · synthetic sample documents
14
Indian document types
0.15 s
Per card on a laptop GPU
100%
Fields read on the test set*
160
Automated tests in CI
*552 fields across 107 synthetic documents, including blurred, rotated, compressed and phone-photo copies, with no wrong field left unflagged. Real-world accuracy is measured per customer in a pilot.
How the parts fit together
One request, and a bulk job
Extract
Jobs
Tech stack
A Python FastAPI service around free OCR models: RapidOCR on the GPU for speed, and PaddleOCR-VL on vLLM for hard images. Postgres holds keys, usage and the job queue, nginx spreads load over replicas, and one Docker Compose command runs it all on a lender’s own servers.
Inner ring: core API · Outer ring: OCR models, infrastructure, monitoring, CI
Build history · real commits
$ git log --graph main
5aa2586
feat(validate): Checksum validators and FastAPI skeleton
13631df
perf(ocr): RapidOCR on the GPU, about 8× faster
e5feaf7
feat(vl): PaddleOCR-VL accurate tier served by vLLM
0947662
feat(docs): Extractors for 14 documents and statement analysis
8e0cf5d
feat(auth): API keys, rate limits, quotas, usage metering
73e07d6
test(eval): Accuracy harness on blurred, rotated, dim copies
4ad4025
feat(scale): Backpressure, batch, nginx load balancer
bfb6b8b
feat(jobs): Background jobs, signed webhooks, monitoring
160 tests · CI passingWhere I started
Learning projects

2023 · Learning project · MERN stack
My first full-stack project, built while learning the MERN stack.
A Q&A platform where users can sign up, then post, filter, upvote, comment on and answer questions on any topic.
Collected 00 / 06
Skill Safari
Nandha Arts and Science College
Skill Safari
Microsoft and LinkedIn
01 Now
Nov – Dec 2026
02 Next
Jan – Mar 2027
03 Then
Apr – Jun 2027
04 Later
Jul – Sep 2027
Phase 01 · Now · Nov – Dec 2026
Go concurrency in depth
01Goroutines, channels, worker pools, context cancellation and pprof profiling.
Transactions & isolation levels
02ACID, row locks and optimistic locking so balances never drift.
Double-entry ledger design
03Record every rupee as matching debits and credits, with a full audit trail.
Open to work: on-site, hybrid or remote
Shipping lending software for
012345678901234567890123456789days
Elsewhere
LinkedIn ↗Send a message