SVC/0x01 Identity

Sahil Patani

Backend Engineer / Bangalore, IN / 2.5 yrs uptime

I build and operate the systems that move money, schedule people, and stay up.

Inspect System Open Channel Download CV

ingress · replay 0 rps

Uptime 2y 6mo

Requests / mo 0

Users reached 0k+

Open incidents 0 4 resolved · 0 regressions

02 Readme who is operating this

Most of what I build is invisible when it works.

Payments that settle once. Schedules that don’t double-book. Migrations nobody notices.

Two and a half years operating production systems end‑to‑end — not just shipping features, but owning them after they ship. That means designing the schema, writing the API, and then being the one who reads the replica at 2am when revenue looks wrong.

The work I’m proudest of is the root‑cause record: defects that had been quietly running for years, traced against live data, blast radius quantified, then fixed properly rather than patched. A double‑charge race across three payment gateways. A pipeline where 40% of calls silently never fired. Things nobody had a dashboard for.

Before that, I was technical point of contact for Uber, HDFC Ergo and InCred — diagnosing production issues inside customer environments, where you don’t get to say “works on my machine”.

operator.json read-only

Role: Backend Engineer — CodeYoung
Base: Bangalore, India · ap-south-1
Focus: Payment integrations, scheduling systems, data migrations, production RCA
Runtime: Node.js · TypeScript · Python · Spring Boot · FastAPI
Storage: MongoDB · PostgreSQL · MySQL · Redis
Prior: Thriving Springs — 200k+ user LMS platform, enterprise POC
Degree: B.Tech Information Technology, Manipal University Jaipur — 8.9/10
Algorithms: 300+ problems solved on LeetCode
Status: Open to opportunities

03 Stack hover a node to trace its path

Client

Edge

Service

Async

Data

mesh Five tiers, twenty-two components. Hover any node to light the path it actually takes in production.

Cloud & DevOps

AWS EC2S3SQSIAMRDS DockerKubernetesLinuxCI/CD

Observability & Reliability

SentryWinstonLoggly JestProduction RCAIncident Investigation

Integrations

StripePayPalIn-house Gateway ZoomWhatsApp Business APICRM Migration

04 Deployments 5 featured · 6 additional

D-01 loop — scheduling engine codeyoung In production

An SQS-backed scheduling engine that never bills twice

I own the student‑reschedule and mentor‑substitution flows in Loop, an SQS‑backed engine processing 13k+ requests a month. Shipped a deferred cancellation ledger that eliminated a defect charging students twice for a single cancellation — cancellations are now recorded as intent and settled once at interval close.

Throughput13k+/mo

Flows owned2

Double charges0

Node.jsAWS SQSTypeScriptMongoDBJest

request path

D-02 crm migration — dual-write cutover codeyoung Lead

Replacing a CRM under live traffic, with a rollback that needs no deploy

Leading a zero‑downtime CRM migration across two backend services. The design is a config‑driven adapter layer: both CRMs sit behind one interface, so dual‑write, phased cutover and instant rollback are all config changes — no redeploy, no traffic drain. 135k+ operations synced across 9 flows since rollout.

Ops synced135k+

Flows9

Downtime0s

Node.jsAdapter PatternDual-writeSequelizeFeature Flags

write path

D-03 billing mesh — 3 gateways codeyoung In production

One billing interval, three payment providers, no drift

Extended interval‑based billing across Stripe, PayPal and an in‑house gateway — including custom multi‑month PayPal plans and mid‑cycle vendor switching that preserves interval metadata so a customer moving providers never gets double‑billed or skipped. The same surface later exposed a webhook race that had been double‑charging since 2022.

Gateways3

Defects traced22

Billing drift0

StripePayPalWebhooksIdempotencyNode.js

settlement path

D-04 lms platform — 200k+ users thriving springs Shipped

Backend for an enterprise LMS serving 200k+ concurrent-access users

Designed and owned the backend APIs for LMS and OJT workflows in Node.js and FastAPI (Python), powering automated role assignment and skill mapping for 200k+ active users under concurrent access. Redesigned MongoDB schemas with index‑aware access patterns — 15% lower average query latency. Redis TTL caching held response times under traffic spikes, and 100k+ production records migrated through staged rollouts with integrity checks and backward‑compatible contracts — zero data loss.

Active users200k+

Query latency−15%

Records migrated100k+

Node.jsFastAPIMongoDBRedisDockerKubernetesAWS

read path

D-05 linkedin profile api personal Live

Structured JSON from any LinkedIn profile, with no browser in the data path

Reverse‑engineered LinkedIn’s internal Voyager API to turn a profile URL into structured JSON — experience, education, skills, images — with no browser automation and no HTML parsing anywhere in the data path. Seven calls against the Dash REST surface, a three‑layer cache (seed / memory / Redis), and curl_cffi to match TLS fingerprints. Deployed on a bring‑your‑own‑session model: callers pass their own cookies per request, and nothing is stored or logged.

Headless browsers0

Cache layers3

Sessions stored0

PythonFastAPIPydanticRedisDockercurl_cffi

Live demo Source free tier — first request after a quiet period takes 30–60s to wake

resolve path

04b Also shipped

Reminder pipeline	Rebuilt AI voice‑call, WhatsApp and email outreach as a scheduled‑task system with atomic call reservation and deduplication windows — every reminder auditable.	35k+/mo · 0 dupes
Scheduling-requests API	Re‑architected from per‑mentor rows to request‑level grouping: faceted counts, server‑driven search, filtering, pagination and a mentor‑eligibility endpoint. Validated against live staging data.	Jest covered
RCA toolkit	Read‑only production‑replica investigation CLIs — query, schema and entity‑timeline tools plus incident playbooks — for root‑causing live revenue defects safely.	3 CLIs
Data migration	Migrated and validated production records via staged rollouts, integrity checks and backward‑compatible data contracts.	100k+ · 0 loss
Bulk CSV ingestion	Schema‑validated bulk import with partial‑failure recovery, built for onboarding at 200k+ user scale.	200k+ scale
Dynamic form builder	Layered Spring Boot backend (Controller–Service–Repository) with a normalised schema separating forms, fields, responses and response values — arbitrary form composition, scalable response storage.	React · Spring

05 Incidents 4 resolved · company & client names withheld

Anyone can list what they built. This is what I found already broken.

Blast radius

22 recurring customer cases

3 payment gateways affected

2022 earliest occurrence

0 open after fix

Root cause

Concurrent webhook deliveries mutated the subscription ledger with no idempotency guard. Two deliveries for the same provider event both passed the “already charged?” read before either committed its write — a classic check‑then‑act race, invisible under normal load and only reproducible when the provider retried.

Detection

Found by querying a read‑only production replica with purpose‑built entity‑timeline tooling — reconstructing each customer’s charge history end to end rather than trusting aggregate reports. The blast radius was quantified before the fix shipped, not after.

Resolution

Reserve‑then‑write on the ledger with an idempotency key derived from the provider event id, making the handler replay‑safe. Reconciled the historical cases across all three gateways.

Blast radius

2× charge per cancellation

Direct customer-facing revenue

0 recurrence since ledger

Root cause

Cancellation and rebill ran as two separate transactions. A cancellation landing mid‑cycle was counted by both the closing interval and the opening one, so the customer paid for the same period twice.

Resolution

Introduced a deferred cancellation ledger. Cancellations are recorded as intent rather than applied immediately, then settled exactly once when the interval closes — making the operation idempotent regardless of when in the cycle it arrives.

Verification

Replayed historical cancellation sequences against staging with live‑shaped data; covered the settlement boundary with Jest suites so the regression cannot return silently.

Blast radius

40% of feedback calls dropped

0 errors surfaced

35k+ monthly reminders in scope

100% auditable after fix

Root cause

Dispatch relied on a scheduler that read tasks without reserving them. Overlapping execution windows caused tasks to be skipped, and because a dropped task threw no error and had no dead‑letter path, the failure was completely silent — the dashboards showed success.

Resolution

Rebuilt the pipeline as a scheduled‑task system with atomic call reservation and deduplication windows across all three channels — AI voice, WhatsApp and email — eliminating both drops and duplicate outreach.

Follow-through

Every reminder now writes an audit record, so “did this fire?” is a query rather than an assumption. The failure mode is observable by construction.

Blast radius

95 sessions misrecorded

Downstream reporting corrupted

95 backfilled correctly

Root cause

Attendance was derived from a single webhook that could legitimately arrive before the session record existed. Those early events matched nothing and were discarded rather than queued — so a genuinely attended session was recorded as a no‑show.

Resolution

Buffered late and early events, then reconciled attendance against the full entity timeline instead of a single signal. Backfilled all affected sessions.

Lesson

Any system deriving state from one webhook is one ordering assumption away from being wrong. Event order is a guarantee you have to build, not one you receive.

06 Changelog newest first

v2.0.0 — current

Software Engineer

CodeYoung · Bangalore March 2026 – present

Node.jsTypeScriptAWS SQSStripePayPalSequelizeJest

+Lead a zero-downtime CRM migration (Cratio – Arali) across two backend services, designing a config-driven adapter layer enabling dual-write, phased cutover and instant rollback without a deploy — 135k+ operations synced across 9 flows.

+Own student-reschedule and mentor-substitution flows in Loop, an SQS-backed scheduling engine processing 13k+ requests/month; shipped a deferred cancellation ledger that eliminated a double-charge defect.

~Re-architected the admin scheduling-requests API from per-mentor rows to request-level grouping — faceted counts, server-driven search, filtering, pagination and a mentor-eligibility endpoint. Validated against live staging data, covered by Jest suites.

~Rebuilt the demo reminder pipeline (AI voice calls, WhatsApp, email) as a scheduled-task system with atomic call reservation and dedup windows — eliminating duplicate outreach across 35k+ reminders/month and making every reminder auditable.

+Extended interval-based billing across three payment gateways (Stripe, PayPal, in-house), adding custom multi-month PayPal plans and mid-cycle vendor switching that preserves interval metadata to prevent billing drift.

+Built a read-only production-replica investigation toolkit (query, schema and entity-timeline CLIs with incident playbooks) and used it to trace a webhook race that double-charged customers (22 cases across 3 gateways since 2022), prove 40% of AI feedback calls were silently never firing, and surface 95 demo sessions wrongly marked unattended.

v1.0.0

Software Engineer

Thriving Springs · Hyderabad February 2024 – February 2026

Node.jsFastAPIMongoDBRedisAngularDockerKubernetesAWS

+Technical point of contact for enterprise clients including Uber, HDFC Ergo and InCred — diagnosed production issues, coordinated fixes and supported live feature rollouts inside customer environments.

~Redesigned MongoDB schemas with index-aware access patterns and consolidated duplicated data paths — cutting average query latency 15% and reducing stored-data redundancy.

+Designed and owned backend APIs for LMS and OJT workflows in Node.js and FastAPI (Python), powering automated role assignment and skill mapping for 200k+ active users under concurrent access.

~Migrated and validated 100k+ production records via staged rollouts, integrity checks and backward-compatible data contracts with zero data loss; built bulk CSV ingestion with schema validation and partial-failure recovery for 200k+-user onboarding.

~Stabilised API response times under traffic spikes using Redis caching with TTL-driven invalidation; deployed containerised microservices on Docker and Kubernetes in AWS-hosted environments.

v0.1.0

B.Tech, Information Technology

Manipal University Jaipur October 2020 – July 2024

DSASystem DesignOOPNetworks

+Graduated with a CGPA of 8.9 / 10.0.

+Solved 300+ data structures and algorithms problems on LeetCode.

07 Connect type help or resume

sahil@console — zsh interactive

›

Hiring, or just want to argue about idempotency?

Email spatani9@gmail.com Phone +91 93095 50866 GitHub sahilpatani1012 LinkedIn sahil-patani LeetCode spatani9 · 300+ solved Résumé Sahil-Patani-Backend-Engineer.pdf · 1 page

Location Bangalore, India · UTC+5:30