Fintech · public data · privacy-first web apps · applied AI with guardrails
🔍 LoanLensCheck a loan offer against RBI's rules before you sign.
Recomputes the true APR from a loan's Key Facts Statement using RBI's method (it reproduces RBI's own worked example, 17.07%). It flags the malpractices RBI has named, such as interest charged before disbursal, advance EMIs with interest on the full amount, and illegal pre-payment charges, and each finding cites the circular it relies on.
|
Find the bank charges you can get reversed.
Reads an Indian savings statement, matches every charge to its GST line, and tests it against RBI rules: minimum-balance penalties that pushed the balance negative, ATM fees within the free limit or above the ₹23 cap, online NEFT charges. Each claim shows the statement lines that prove it, and it drafts the dispute. Nothing leaves your browser.
|
|
Indian health-insurance data, explained without guesswork. Open public-data explorer built on official IRDAI handbook tables. Every metric carries its source, year and formula. Missing data stays missing instead of being guessed. Automated monthly source checks open review issues.
|
📈 IPO LabUnderstand the issue. Research the issuer. Follow the proceeds. Evidence-led IPO research platform that links SEBI/NSE/BSE offer documents to fundamentals and tracks how IPO proceeds were actually used, with page-level citations. Includes 27 learning modules and 6 interactive tools.
|
🤖 RecoverX: an AI revenue-recovery agent with hard guardrailsAn LLM reads customer replies to understand intent, then deterministic policy rules decide what happens: retry, payment link, follow-up, escalate or stop. Disputes, fraud and high-value cases can never be auto-actioned. On a 47-transaction benchmark, it matched rules-only recovery with lower automated exposure (48.8% vs 42.7% attempt success). It also keeps a full audit trail. |
|
| Repo | What it shows |
|---|---|
| Booking-System- | Generic ring-buffer queue. Fixed a wrap-around bug and a never-reusable-slot bug, each with a regression test. |
| Flight & Train booking | Leak-free unique_ptr linked queue and stack with iterative teardown, safe for 1M+ nodes. |
Both are built in CI with g++ and clang++ under -Werror, then tested again under AddressSanitizer and UBSan.
- Evidence over vibes. Every number links to a source, a formula or a test. "Not available" beats a guess.
- Privacy by construction. If a tool doesn't need your data on a server, it never sends it there.
- AI with guardrails. Models interpret context; deterministic rules make the decisions that matter.
- Ship it properly. CI on every push, CodeQL scanning, tests that pin real bugs, and live demos.
Open to internships and collaborations in fintech, data and applied AI. Open an issue on any repo to reach me.