Real projects, real business impact — each one verified by production data.
Crisis Management
Recovered 137,000 lost financial records — with zero left behind
✓ 100% of data restored · damage stopped same day · permanent fix shipped
For 7 days, the company's pricing system silently corrupted profit-and-loss data on
~137,000 orders — and no alarm ever went off. Though I hadn't caused the issue,
I took ownership of solving it.
- Diagnosed a failure so deep in the system that it slipped past every safety net — the kind of bug most engineers never find
- Stopped further data loss the same day the problem was discovered
- Wrote and ran a recovery program that restored all 137,000 records, independently verified by the Data Science team
- Delivered the permanent fix, wrote reports for both business and engineering leadership, and mentored the developer involved
Root-cause analysisData recoveryJavaIncident reports
Systems Design
Designed and built an event-driven Payout System from scratch
✓ 30,000+ customers paid out daily · idempotent, DLQ-backed retries
Wallet-routed payouts across marketplaces had no dedicated system. I owned it end to end —
from HLD/LLD through production rollout, with no separate architect and no hand-off.
- Independently authored the high-level and low-level design, then shipped production-ready code myself
- Built an event-driven pipeline (Airflow cron → SQS → Lambda) that's idempotent by design, with DLQ-backed retries so one failed payout never blocks the batch
- Routes payouts by marketplace and wallet, serving 30,000+ customers every day
System designAWS LambdaSQSAirflow
Revenue Impact
Redesigned the home page into a revenue-driving surface
✓ ₹10–13L in daily GMV · 40% lift in Gold Membership renewals
The customer app's home page was static. I rebuilt it as a set of targeted, time-slotted
surfaces that drive measurable revenue, not just impressions.
- Built a customer-specific, time-slotted Hero Banner promoting brand products — driving ₹10–13 lakh in daily GMV
- Added Merch Banners and a Delivery-vs-Credit "Nudge" card to the same surface
- Extended the Nudge mechanism to Gold Membership renewals, lifting the renewal rate by 40%
JavaPersonalizationGrowth
AI Engineering
Built a harness of AI agents that do real engineering work — not autocomplete
✓ Autonomous SDLC pipeline · knowledge-graph-backed team agent · zero-ticket ops tooling
Alongside my day-to-day engineering, I design and build agentic AI systems — real production
tooling that takes over engineering and operations work end to end.
- A multi-agent AI harness that takes a work ticket through requirements, architecture design, low-level design, implementation, an adversarial multi-round code review, and automated testing — with zero human intervention on routine tickets. It checkpoints and resumes across interruptions, and pauses for a human-in-the-loop decision when a change has real architectural stakes.
- An AI agent that acts as an on-call virtual team member — it uses tools to query a knowledge graph of our code, business logic, and past decisions, reasoning over the graph rather than just retrieving text to answer engineering and business questions. The same graph feeds the harness above, so every run makes the next one smarter.
- An AI-assisted incident-recovery agent that automates data-repair jobs with automatic retry on failure — it recovered tens of thousands of records during the 137K-record crisis above.
- An AI documentation agent that writes and publishes root-cause-analysis reports straight to the team wiki, plus a self-service diagnostic tool that lets support staff resolve customer payout disputes without ever filing an engineering ticket.
Agent harness designMulti-agent systemsKnowledge graphsLLM tooling
Building From Scratch
Designed company-critical infrastructure in 7 weeks
✓ 190K requests/month · unblocked every dependent team
A mapping system the whole field-sales product depended on existed only as an idea —
no requirements, no design, nothing. I turned it into working infrastructure.
- Wrote the product requirements and technical design documents myself
- Built it across 6 backend services
- Every team waiting on this work was unblocked; it now handles ~190,000 requests a month
System designJavaGeo/mappingDocumentation
Product Launch
Launched a paid membership product that's still growing
✓ 11+ features shipped · ~9,000 customer interactions/month a year later
Jumbotail's premium "Gold Membership" — a new revenue line for the company.
I built the backend features customers actually touch.
- Savings tracker that shows members the real money they've saved — making the membership's value visible
- Reward experiences (scratch cards) that make joining feel exciting
- Subscription-management APIs — cancellation, expiry status, backend-driven renewal messaging — running the membership's full lifecycle
- Automated WhatsApp and app notifications for renewals and expiry — no manual work needed to retain members
- Gave the operations team self-serve tools, removing their dependency on engineers
JavaProduct engineeringWhatsApp automation
Platform
Built the Communication System every team now publishes through
✓ 5,000+ messages/day · powers OTP + 6 lifecycle automations
Teams needed to send WhatsApp, SMS, and push messages with no shared, reliable way to do it.
I designed the shared platform everyone routes through now.
- Any team publishes a message event to a dedicated SQS queue; the system routes it through the right channel (WhatsApp/SMS/PN)
- Powers OTP delivery for login, delivery, and credit flows, plus 6 event-triggered lifecycle automations
- Sends 5,000+ messages daily in production
SQSWhatsApp/SMS automationMulti-tenant
Partnership Integration
Made a two-company partnership actually work
✓ Eliminated an entire category of customer-facing bugs
Jumbotail's platform serves a partner marketplace (SOLV), whose customers kept
getting the wrong offers, messages, and support. I fixed it — permanently.
- Instead of patching each bug, I found the single root cause affecting every partner customer's login — and eliminated the entire bug class
- Designed a secure login system bridging the partner's identity platform (AWS Cognito) with ours, built so partner outages never block customers from logging in
- Built a rollout tracker giving field reps a distance-sorted list and urgency signal for which customers still need to migrate to the partner's new app — turning a blind rollout into a targeted one
- Routed partner support tickets to the right teams automatically
Security/login systemsAWSSupport automation
Support Tooling
Built the in-app ticketing system the entire sales team now uses
✓ 15% faster issue resolution · used by the entire field-sales team
Field reps had no direct way to report a customer issue on-site. I built the pipeline
end to end, then kept iterating on it as new needs came up.
- End-to-end pipeline from the mobile app through the API gateway to a ticketing service, routed into Freshdesk
- Iterated it through a dozen+ follow-on improvements: multi-image attachments, dynamic issue categories, marketplace-based routing
- Now used by the entire sales team, resolving issues 15% faster
JavaFreshdeskMobile-to-backend
Initiative
Fixed failures nobody asked me to fix
✓ Found & fixed a 3-year-old failure the same day I spotted it
I monitor production health on my own initiative — and act on what I find.
- Spotted a feature that had been failing 93% of the time for 3+ years — traced three separate bugs and had the fix approved the same day
- Resolved a login outage hitting users every 7–16 minutes — diagnosed, fixed, tested, and documented in 2 days
- Fixed a map-display bug blocking a app feature within 1 day of it being reported
MonitoringProduction debuggingFast turnaround
Building From Scratch
Built a pipeline that feeds real market prices into pricing decisions
✓ Direct app-to-S3 invoice capture · automated extraction
The pricing team needed real competitor price signals with no manual data entry.
I built the capture pipeline from scratch.
- Backend issues presigned S3 URLs so the app can upload invoice photos directly to S3 — no server-side file handling
- An S3-triggered Lambda extracts invoice data automatically via API
- Feeds market-price signals the Pricing team uses to adjust product prices
AWS S3LambdaPresigned URLs
Systems Ownership
Owned the return-policy system end to end
✓ Geo-specific policies, reliably propagated
Return windows vary by geography and change over time — customers need to see the
right one, every time.
- Built geo-specific return-policy ingestion
- Async propagation via a versioned SQS queue with DLQ-backed retries
- A UINode-based display layer shows the correct return window on the product page
SQSUINodeJava