BCS Rohith

I turn live event streams into decisions that ship.

I lead the AI Platform team at RudderStack, building an agentic decisioning product end to end. Before that, five years on the data plane underneath it: pipelines, queues and warehouses that move customer events at scale.

6.7TEvents processed through Postgres queues

100k/sPostgres queue throughput

~80%Lower warehouse load cost

40%Faster partition rebalancing

Experience

RudderStack ↗

Hyderabad · Dec 2020 – Present

Tech Lead, AI Platform
Aug 2025 – Present
- Own architecture, roadmap and delivery of an agentic decisioning product, from prototype to production.
- Ship to production weekly across server and UI; unblock enterprise rollouts with product, sales and legal.
- Led the LangChain to Mastra runtime migration, shadow mode and the platform's MCP tooling.
- TypeScript
- Go
- Postgres
- ClickHouse
- Temporal
- Mastra
- MCP
Software Development Engineer 3
Dec 2020 – Aug 2025
- Co-designed the multi-tenant Rudder server for Pro and Growth tier customers.
- Built a unified dead letter queue with replay; cut partition rebalancing time by 40%.
- Reduced warehouse load cost by about 80%; regionalised shared services for GDPR.
- Go
- Kubernetes
- Postgres
- ScyllaDB
- BadgerDB
- Helm

MyAlly.ai

Hyderabad · Jul 2019 – Dec 2020

Software Engineer
Jul 2019 – Dec 2020
- Sped up the scheduler for complex onsite interview loops.
- Prevented interviewer double-booking under race conditions with Redis locks.
- Python
- Django
- Redis
- AWS Lambda
- SQS

IBM India ↗

Bengaluru · Jun 2018 – Jul 2019

Data Scientist
Jun 2018 – Jul 2019
- Built energy-efficiency analytics and a real-time insights chatbot for retail energy managers.
- Python
- Cloudant
- Db2

Selected work

AI Platform

Agent runtime swap

Moved every live agent from LangChain to Mastra, with both running behind one switch.

0downtime for live agents

AI Platform

Shadow mode

Agents decide on real traffic with every send held back, so teams can trust them before going live.

100%of sends held back until go-live

Data plane

Postgres as a queue

Event queues on plain PostgreSQL, tuned for sustained high throughput in production.

6.7Tevents processed

Data plane

Unified dead letter queue

One home for every event that fails delivery, with replay back to its destination.

1framework for every failure path

Talks & recognition

PGConf India · 2026 · Watch ↗

Hard-won lessons after processing 6.7T events through Postgres queues

Hyderabad PostgreSQL User Group · Watch ↗

0 to 6.7T: Real-World Lessons Scaling PostgreSQL as a High-Performance Queue

RudderStack Hackathon · 2026

Runner-up, leading Team Voyage

Compiled plain-English marketing goals into live campaigns.

IIT Madras · Dual Degree, Electrical Engineering2013 – 2018
CAT · 99.92 percentile (2018), 99.62 (2017)2018
JEE · top 1% Advanced, top 0.1% Main2013

Have an agent or a pipeline that has to work in production?

Let's talk. I'm happy to compare notes on agent platforms, Postgres queues, or speak at your meetup.

LinkedIn ↗ GitHub ↗

rohith.bcs@gmail.com