SOFTWARE ENGINEER
Hrishikesh
Bhardwaj.
4.5 years building high-throughput distributed systems at ION Group, from FX execution to post-trade processing. Now a Master of Software Engineering student at Carnegie Mellon, focused on reliable backends, cloud infrastructure, and applied AI.
Available January 2027
hrishikb@andrew.cmu.eduLinkedIn ↗
A look inside the work. Choose a project, explore the code, or read the story.
Applied AI · CMU capstone
Attribute Prediction
Catalog ingestion and attribute prediction for eParts Services LLC, built by the five-engineer team I led as forward deployed engineer.
FOLLOW THE FLOW
Normalize supplier catalog records into one attribute schema.
Route per attribute rather than per record. Each prediction carries its own confidence, so a reviewer sees only the fields the system is unsure about, and their corrections feed retraining.
Manual catalog review dropped three to five times against the process the client had been using, across ingestion, prediction, routing, and writeback.
Client-owned data and deployment; measured on their catalogs rather than a public benchmark, and the source stays private.
NOTES FROM BUILDING
From the blog
Course and client work. The source is private; these are my own notes on building it.
All blogs ↗I built a Twitter analytics service. Most of the work was figuring out why it was slow.
20,000 requests per second at $0.31/hour in Phase 3, plus a separate 70,000-RPS Redis-only experiment that traded cost for throughput.
Uber Ride Matching: making the next event count
Inside my Kafka and Samza ride-matching backend: partitioning by city block, tracking driver availability, and joining live events with rider profiles.
One chat application, two clouds, one deployment model
I took supplied Spring services through containers, Helm, GKE, and AKS, making the differences between clouds explicit while keeping one understandable application definition.