VA

Venkhatesh Arunachalam

ML & Data Platforms • Retrieval & Ranking • Low‑latency Serving

I design pragmatic, measurable ML systems: streaming ingestion → features → retrieval/ranking → GPU serving. I optimize for latency, reliability, and relevance—keeping systems simple and observable.

Email GitHub LinkedIn 1‑page resume

What I do

  • Own end‑to‑end ML/data backends for search, recommendations, and analytics.
  • Ship retrieval + ranking systems (dense vectors + BM25) with measurable relevance lifts.
  • Reduce inference cost/latency with ONNX Runtime, FAISS, semantic caches, profiling.
  • Build streaming ETL (Kafka/Flink/Spark), batch pipelines (Airflow/DBT), and observability.

Highlights

  • Clinical search serving 12k+ users; p95 < 150ms at ~1k QPS.
  • Hybrid retrieval + re‑ranking improved NDCG@10 by 22%.
  • CUDA/PyTorch profiling cut LLM latency & token costs by ~40%.
  • Streaming ETL: ~245GB/day with strong fault‑tolerance.
  • Cut reporting latency 30 min ➜ 2 s; 60% CPU reduction on PO pipeline.

Recent roles

  • ML Engineer, Fidari Care — retrieval & clinical search; real‑time data ingests; GPU/ONNX serving.
  • Research SWE, Indiana University — multi‑university mental‑health data platform; streaming ETL.
  • ML Eng Intern, FourKites — distributed PO pipelines, Elasticsearch sync, observability.
  • Software Engineer, Fynd — high‑performance APIs & data infra; DBT/Spark/Redshift.

Awards

  • Winner — NVIDIA–Atos–CDAC National AI Hackathon.
  • 2nd Place — HERE Maps Hackathon.
  • Fynd Star Award; SIES GST Technical Excellence (2020).

Stack

Languages: Python, C/C++, Java, Go, JS/TS

ML & Data: PyTorch, TensorFlow, ONNX Runtime, Spark, FAISS, Elasticsearch, Ray, Triton

Infra: Kafka, Flink, Airflow, Kubernetes, gRPC, Prometheus/Grafana

Cloud & DB: AWS, GCP, Redshift, Postgres, MySQL, MongoDB

Education

M.S. Data Science — Indiana University Bloomington (2024)

B.E. Computer Engineering — Mumbai University (2020)