Aryan Verma

About Aryan Verma

I’m a systems-oriented software engineer with experience building large-scale, high-performance platforms in feed infrastructure and MLOps. My work has spanned from developing low-latency, high-throughput search engines in Rust to designing observability systems that support ML models across production at LinkedIn.At LinkedIn, I’ve contributed to the migration of core IR systems to a Rust-based engine (handling 1.2M QPS), built an embedded RocksDB-backed feature store, integrating them to both our Java stack and Kubernetes container stack and migrated old Java filters and scoring models to be executed in our engine itself as a form of query. I’ve also worked with the Health Assurance team on AI Studio, building monitoring APIs and systems used to track hundreds of ML models.Previously at Amazon, I designed and scaled our video transcoding platform for all Ads videos, created our spec fix service to automatically detect and transform videos to be usable across every ad program and created a backfill platform to do backfills for all ingested videos whenever a new ad program was introduced.I enjoy working at the intersection of infra, ML systems, and performance engineering, and I’m passionate about building reliable platforms that scale.Let’s connect if you’re interested in distributed systems, ML infra, or building for scale.

Experience

1. Software Engineer

   Linkedin

   Apr 2024 — Present · Bangalore Urban, IN

   Information Retrieval Platform (L0 Retrieval/Recommendation Engine for Linkedin’s Feed)- Contributed to the design and development of Linkedin’s high-performance, Rust based retrievalengine for its Feed L0 stack, built to handle 1.2 Million QPS- Built and optimized read-path API’s for a custom RocksDB-backed feature store supporting multiplelow-latency reads per query- Integrated multiple feature stores with index freshness/integrity checks and parallel downloads in a Kubernetes environment- Wrote the translation of multiple filter policies(filtering of spam, inappropriate content etc) from java to our in-house query language built for NemoDB- Migrated old ML models for L0 scoring that was written in Quasar(in house inference framework) toNemoDB query and wrote integration tests.

Education

• Delhi Technological University (Formerly DCE)

  Bachelor's degree

  2017 — 2021