Skip to content
View karthy1004's full-sized avatar
1 follower · 1 following

Block or report karthy1004

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
karthy1004/README.md
Karthik Bhumireddy. Product leader in payments and responsible-gold due diligence, who also builds the systems: state machines, audit trails, evidence-backed decisions. Principal PM at Qenta, previously Paytm and Dineout, co-founder of Deccan Foods, MBA from Rice.

LinkedIn · Featured work · Skills, with evidence

I've spent my career on the business side of trust: building a food brand as a co-founder, then payments products at Paytm, Dineout and now Qenta, where I lead the Responsible Gold due-diligence product. The question underneath all of it is the same: how do two parties who don't know each other trust a transaction? These repositories are my answers in code. Each one makes a decision and keeps the evidence for it.

Featured work

payee-verdict: one payee-name decision across four payment systems. EU VoP, UK CoP, India and PIX, simulated. 181 hard cases: 96.1 percent, zero false matches. 143 tests and a replayable evidence log. Python. Working demo.
payee-verdict CI status
One payout decision across four name-check systems. Zero false matches on 181 hard cases.
python -m payee_verdict demo 		payments-reliability-lab: payments that stay correct when everything retries. A timeout is unknown, never failed. Exactly-once webhooks, one charge per retry. 59 tests; all five reconciliation break types tested. TypeScript. Working demo.
payments-reliability-lab CI status
Timeout means unknown, never failed. One charge however often the client retries.
npm run demo
evidence-review-engine: due diligence that scores evidence, not claims. 12 yes answers with bad documents score 16 out of 100. LLM extraction gated by grounding checks. 59 tests, 98 percent branch coverage. Python. Working demo.
evidence-review-engine CI status
Scores evidence, not claims. 12 "yes" answers with bad documents score 16/100.
python -m evidence_review demo 		operon: job applications done for you, you keep the veto. Exactly-once Stripe webhooks in test mode. Approval engine with compare-and-set. 20-model domain and 13 unit tests. Next.js, Prisma, Stripe. Prototype, not launched.
operon CI status
Prototype, not launched. Exactly-once Stripe webhooks and a race-safe approval engine.
Read-only demo: running it needs Clerk, Stripe and Postgres

Every number on these cards comes from a real run of the linked repository, and each README shows the output. All four are clean-room builds with synthetic data: no employer code, data or methodology. Each README also says who the product is for, what I'd measure, and what I cut.

How I work

  • Claims need evidence. In the due-diligence engine, an unbacked "yes" scores zero. My skills matrix applies the same rule to me.
  • Design the failure path first. A payment timeout means unknown, not failed. Getting that wrong is how customers get charged twice. (ADR-1)
  • Optimise for the expensive mistake. payee-verdict must have zero false matches before accuracy counts, because a wrong "match" sends money to the wrong person. (corpus results)
  • AI only where it can be measured against a baseline. The LLM step must pass schema, consistency and grounding checks, and it's scored on silent errors against a regex baseline. The live-model run is still to come. (eval)
How my work connects
Map: payments and fintech, responsible gold, digital identity and entrepreneurship all lead to one theme, evidence and auditability, which leads to four projects: evidence-review-engine, payments-reliability-lab, payee-verdict and operon.

At Qenta I lead the Responsible Gold due-diligence product and contribute product strategy across its payments, G-Coin (digitised gold) and QID (decentralised identity) work. I build these prototypes with Claude Code as an AI pair programmer, and every claim in them is checked against tests and real runs.

No stats widgets or visitor counters on purpose: the profile's images are generated from one script, and CI checks every link and image.

Popular repositories Loading

  1. evidence-review-engine evidence-review-engine Public

    Explainable due-diligence scoring: versioned rules, evidence-backed answers, audited review workflow, and an LLM extraction step gated by grounding checks and evals.

    Python

  2. payments-reliability-lab payments-reliability-lab Public

    A payment API that stays correct under timeouts, retries and duplicate webhooks: idempotency keys, HMAC webhooks, exactly-once events, reconciliation. Simulated provider.

    TypeScript

  3. operon operon Public

    Managed job-application service prototype: Next.js + Prisma domain model, Stripe test-mode checkout with exactly-once webhooks, customer approval engine.

    TypeScript

  4. karthy1004 karthy1004 Public

    Profile README

    Python

  5. payee-verdict payee-verdict Public

    Explainable payee-name verification across EU VoP, UK CoP, India and PIX name checks: reference matcher, versioned proceed/warn/hold policy, hash-chained evidence log with replay. Synthetic data.

    Python