Evidence ledger · updated 02 Nov 2026
Every claim has a receipt.
A claim is only useful if you can check it. Each row below links to a live build, a public repository or a case record. Where the evidence is private, the row says so instead of dressing it up.
01 / The ledger
Claims and sources
Each row is tagged by how you can check it: measured from a build or test, public and directly verifiable, or private because employer code is not mine to publish.
- measuredProduced by a build, test or instrumented run.
- publicA live deployment or repository you can open yourself.
- privateEmployer work, described to the limit of the NDA.
SwasthyaSetu writes to IndexedDB first, then drains an idempotent EVT- queue. Verified through local partition and reconnect scenarios.
A tested behaviour, not a production SLA.
MediVault's break-glass QR pass exposes only blood group, allergies and vitals. Measured on the deployed build.
Production analytics recorded by Student Placement Prediction during the reported period. Not a projection.
Analytics period is shown in the case study.
An asynchronous thumbnail pipeline built during my SDE internship at Chitralai. Measured in production; implementation is private.
Employer code is not public.
IDOR and XSS fixes at Chitralai, including authorization checks and output encoding. Details are under NDA.
No confidential details are published.
Clean & Green Technology track, April 2026, with EcoVision Bharat: a YOLOv8 civic waste platform.
Three deep case studies and six catalogue builds, with live deployments or public source where available.
02 / Method
How to verify this site
The portfolio is built to be checked, not just read. The lab runs real browser storage and a scoped service worker; nothing is faked with a screenshot.
- Open the labCut the network, queue an intake, reconnect and watch the idempotent outbox drain. Reload the page while offline to confirm the service worker is actually serving it. Open /lab
- Read the case recordsEach case separates the constraint, my ownership, the decisions and the evidence. The reflection section names what broke. Open /work
- Check the sourcePublic repositories are linked directly from every case study. MediVault and SwasthyaSetu are live builds with public source.
- Read the notesThe engineering patterns behind the records, including the failure modes I would fix next. Open /notes
03 / Limits
What I do not claim
The boundary of a claim is part of the claim. These are the limits I would state on a call, written down here first.
- I do not claim a production SLA. The metrics here come from deployed builds and local test scenarios, not a contractual uptime guarantee.
- I do not present employer work as public code. Chitralai and Allkane work is described at the level the NDA allows; the architecture and measured outcomes are fair to discuss on a call.
- I do not attribute team outcomes to myself alone. Each case study lists what I specifically owned.
- I do not claim scale I have not operated. The student analytics are real for the reported period, not a growth projection.
- I do not claim the AI is always right. The pipeline is designed to fail over, flag uncertainty and keep a human in the loop.