Chaitanya Arora

Software Development Engineer III, Enterprise Payments · contract · Remote / Delhi NCR

Backend services for the enterprise payments platform behind GoTyme Bank in the Philippines and TymeBank in South Africa: accounts, pay-ins and payouts for business clients. Unlike consumer banking, where activity spreads across millions of accounts, a few business accounts take heavy concurrent writes, so the system is OLTP and has to stay consistent under contention.

• Built the balance-ledger service from an empty repository to production.
• Designed the payout service's state machine, written up in a decision record the team adopted, and made message redelivery safe to retry.
• Traced a latency collapse in the Java payout service to a synchronized logger rather than the database, and fixed it with asynchronous logging.
• Ran the benchmarks behind the team's datastore decisions, for the ledger and for transaction search.
• Now building the backend for the client portal.
• Rewrote the team's logging standard and gave technical sessions to the engineers.

• Python
• FastAPI
• Java (Spring Boot)
• PostgreSQL
• DynamoDB
• SQS
• EventBridge
• ECS Fargate
• event-driven services

Software Engineer II, lending platform · contract · Remote

A Philippine fintech that offers salary loans to employees through their employers.

• Owned the document-processing service end to end, from proof of concept to production. It reads payslips, government IDs and employment certificates and pre-fills the data for loan officers, cutting manual processing per application from about 15 minutes to about 2.
• Moved its extraction from OCR templates to LLMs, and made the provider and model per document type a configuration change rather than a release.
• Co-designed and co-built the credit-risk service from scratch with one other engineer. Credit decisions went from 8–9 seconds to under 500 ms, and lending-policy changes went from months of engineering to runtime configuration.
• Created its eligibility-validator framework, still in use today, and the employer-policy admin APIs, which are unchanged since I left.
• Wrote the effective-interest-rate (XIRR) compliance logic that keeps every loan offer under a monthly cap. Its core calculation became part of the company's shared SDK.
• Productionised the data-science team's loan-scoring models as a service on AWS.
• Built Tracker Service, unprompted and with one other engineer: a FastAPI and MongoDB app with Google sign-in and two-way BigQuery sync, replacing the spreadsheets collections and finance were working in. It has run for four years with no code changes.
• Wrote the scheduled job that purged the risk service's never-cleaned simulation-log table, cutting it by about 90% and keeping a rolling three-month window.
• Mentored the engineer who took over the risk service through a year-long handover, and ran a Datadog session for the engineering team.

• Python
• Django
• FastAPI
• PostgreSQL
• AWS (Textract, Bedrock)
• Redis Streams
• BigQuery
• MongoDB
• LLMs

Deputy Manager, Information Security (joined as IT trainee) · Delhi NCR

In my final year of college I found a security flaw in their systems and disclosed it responsibly. Two weeks later they hired me, and I was promoted twice in 14 months.

• Built a computer-vision pipeline that automated onboarding for phlebotomists: OpenCV to correct skew and lighting, Google Vision to pull text and faces from ID documents and certificates, and a parser that filled in the onboarding forms.
• Ran the vulnerability assessment and penetration testing programme across applications, infrastructure and medical instruments, and designed the vulnerability-management process.
• Wrote the secure-coding guidelines, and assessed software releases, vendors and the data-lake architecture.
• Built application-security dashboards for senior management, presented weekly to the CEO and CTO and monthly to the managing director, and worked directly with the CISO.

• Python
• OpenCV
• Google Vision
• application security
• VAPT

Independent product, licensed to PVR INOX · sole developer

• Built, and still run, the vendor-invoice portal PVR INOX's accounts payable works through. Vendors submit invoices against purchase orders and track them through approval and goods receipt to payment.
• Built an eight-stage invoice lifecycle and a separate vendor-advance state machine, approval chains that route each invoice by amount, and purchase-order balance, duplicate and GSTIN checks at submission, with a full audit trail.
• Added role-based access for vendors and five kinds of employee, scoped by location and department. Master data, approval workflows and ERP payment files load in bulk by CSV, and one payment can cover several invoices or part of one.
• Moved its ERP integration from Navision to SAP.

• Python
• Django
• PostgreSQL
• S3
• Gunicorn
• Nginx

Technical lead, data and research · an AI legal-tech platform · Remote

• Led the data and research team, three engineers over the engagement plus 4–5 legal interns, that collected India's case law: the Supreme Court, about 25 High Courts, tribunals, GST rulings, central Acts and the e-Gazette. About 18–20 million documents, OCR'd to full text.
• Created the research service and wrote most of its code: full-text search on PostgreSQL and then Elasticsearch, semantic search on Pinecone, and a Neo4j graph of which judgments cite which.
• Re-architected a founder's legal-notice prototype into a multi-tenant product for lenders. It scores defaulted loans and routes each one to soft recovery, a cheque-bounce notice or litigation; uses an LLM to map each lender's templates and borrower files; and sends and tracks notices over WhatsApp, SMS, email and post. It was deployed and in pilot when I handed over.
• Mentored the engineer who now runs the research service, and onboarded my successor.

• Python
• Django
• PostgreSQL
• Elasticsearch
• Pinecone
• Neo4j
• Redis Streams

Built and published on my own

• Hashtag ToolTips shows how many followers a hashtag has, right where you read and write LinkedIn posts. About 1,800 people use it every week, with no marketing. I rebuilt it in 2026 for LinkedIn's current pages.
• My first extension peaked at more than 13,000 weekly users, until changes to LinkedIn's API broke it.

• JavaScript
• Chrome Manifest V3
• 8 languages