Backend engineer at Engati, Bangalore. 2+ years full-time (after a 6-month internship) building Java & Spring Boot services for a high-volume B2B messaging platform.
I build reliable backends for high‑volume messaging.
~/shashank $
~2M api triggers a day · 50k–100k retries. java · spring boot · kafka · redis · mongodb. 45% of failed deliveries saved on 1st retry. open to sde-2 roles · 2026
Open to SDE-2 roles✺Employee of the Month ×2✺MongoDB certified✺CGPA 9.47✺Bangalore✺Open to SDE-2 roles✺Employee of the Month ×2✺MongoDB certified✺CGPA 9.47✺Bangalore✺
Skills: Java, Spring Boot, Apache Kafka, RabbitMQ, Redis, MongoDB, Spark, AWS S3, Microservices. Highlights: Open to SDE-2 roles, Employee of the Month ×2, MongoDB certified, CGPA 9.47, Bangalore.
A production-isolated sandbox built with Nginx rerouting. 40–50 people across engineering, FDE and support test there safely, with zero impact on production. I wrote its configuration, database and backend; the DevOps team scripted the Nginx routing.
(05) Incidents
2 write-ups from production
Production incidents
Production problems I tracked down, written up the way a postmortem would be.
INC-01 · MongoDB M20 · production● resolved
Cluster running out of memory
Impact
Memory exhaustion on the production MongoDB cluster.
Cause
Hot queries had no supporting index: ~3,000 documents scanned for every 5 returned.
Fix
Indexed the high-frequency query fields; scan-to-return fell under 100:1.
−60%memory usage (cut by over 60%)
before100%
after<40%
Before: 100%. After: <40%.
INC-02 · FastAPI · embeddings● resolved
A service quietly leaking memory
Impact
Peak memory climbing to 2.5 GB on a critical service.
Cause
String concatenation inside an embedding loop kept allocating new copies.
Fix
Refactored the loop to in-place operations.
2.5 → 1 GBpeak memory
before2.5 GB
after~1 GB
Before: 2.5 GB. After: ~1 GB.
(06) Playground
interactive simulation · illustrative numbers
The retry flow, simulated
An interactive simulation of my Engati auto-retry framework; its numbers are illustrative, not production data. Triggers flow out to Meta; failures come back as webhooks, through the analytics pipeline, to trigger-mvc. Retryable ones wait in RabbitMQ with back-off, then go out again with their original payload, fetched from MongoDB by trackerId. Switch the framework off, or have Meta send a burst of error webhooks, and watch the failure rate.