Sarabesh Ravindranath

AI/ML Engineer · Bloomington, Indiana

I build reliable machine-learning systems, from experiments and retrieval pipelines to deployment and observability.

Sarabesh Ravindranath

I’m an AI/ML engineer with 4+ years across MLOps, DevOps, computer vision, and retrieval systems. I enjoy taking models beyond notebooks: building repeatable training pipelines, automating delivery, and making production behavior measurable.

Most recently, I’ve been a research assistant at Indiana University’s School of Optometry, where I build reproducible 3D terrain-reconstruction and complexity-modeling workflows and train camera-orientation models on IU supercomputing systems. Before that, I worked on ML training and observability at Musco Lighting and shipped multi-cloud platform automation at VMware and Presidio.

I finished my MS in Data Science at Indiana University Bloomington with a 3.90 GPA. I’m currently interviewing for full-time AI/ML, MLOps, applied scientist, and LLM/RAG roles in the United States, and I’m open to relocating.

Work authorization: F-1 STEM OPT with EAD valid through June 2028. H-1B sponsorship will be required thereafter for continued work authorization or change of status.

You can find my work on GitHub, connect with me on LinkedIn, or reach me by email. You can also download my résumé.

Experience

Research Assistant · Indiana University School of Optometry Jan 2025 — present
  • Collaborate with faculty on the Walking Project, building reproducible 3D terrain-reconstruction and complexity-modeling workflows from data collected across 10+ trails and 8 subjects using RealityCapture.
  • Reconstructed 10+ scenes from binocular and monocular imagery, creating terrain-analysis assets for scene-understanding and surface-complexity experiments.
  • Plan field data collection and evaluate multimodal capture tooling including Meta Aria eye-tracking glasses, drone and overhead cameras, GPS trackers, and Shadow IMU body suits.
  • Train and evaluate single-image camera-orientation models on the Nymeria dataset using IU supercomputing workflows, SLURM job orchestration, environment modules, and distributed training as needed.
Emerging Tech Intern, AI Specialist · Musco Lighting May 2024 — Aug 2024
  • Developed a modular MLOps pipeline with DVC and GitLab CI for dataset versioning, validation, retraining, and model promotion, improving iteration speed by 40%.
  • Implemented YOLOv5 training and batch-inference workflows integrated with MLflow for experiment tracking, artifact management, model comparison, and performance visibility.
  • Designed event-driven dataset synchronization between NFS and S3, reducing data latency by 80% and improving availability for downstream ML workflows.
  • Enhanced production ML observability with MLflow dashboards, drift signals, and data-change alerts to strengthen model lifecycle governance.
DevOps Engineer, MTS II · VMware Feb 2022 — Jun 2023
  • Contributed to the VMware Tanzu Application Platform release team, automating multi-cloud installation and validation pipelines across EKS, AKS, and TKGS.
  • Designed and implemented Go-based end-to-end validation suites, increasing automated test coverage from 20% to 75% across platform-release workflows.
  • Integrated CI/CD workflows into Helm-based Kubernetes infrastructure, improving release reliability and reducing manual validation overhead.
  • Partnered with platform, QA, and release teams to optimize Kubernetes deployments, version-control integration, automated environment validation, and release readiness.
Cloud Engineer · Presidio May 2019 — Jan 2022
  • Architected and deployed enterprise CI/CD pipelines using Jenkins and GitHub Actions, implementing infrastructure as code with Terraform and CloudFormation.
  • Built scalable AWS infrastructure for clients including ViacomCBS, StatsPerform, and WEX, supporting cloud-native deployment, cost optimization, and operational reliability.
  • Led the migration of Jenkins workers to Kubernetes pods, improving resource efficiency and reducing build times by 35%.
  • Developed internal DevOps libraries integrating Slack and Sumo Logic APIs for real-time alerting and monitoring, reducing mean time to recovery.

Skills

Machine learning & AI
PyTorch, TensorFlow, scikit-learn, Hugging Face, LangChain, Open3D, TensorRT, YOLOv5
MLOps & infrastructure
DVC, MLflow, SageMaker, Airflow, Prefect, Argo CD, Docker, Kubernetes, Terraform
Cloud & delivery
AWS Lambda, ECS, EKS, RDS, S3, Step Functions, GitHub Actions, GitLab CI/CD, Jenkins, Helm, Tanzu
Programming & data
Python, Go, C++, Java, Groovy, Flask, Spring Boot, MySQL, DynamoDB, Neo4j, Qdrant
Focus areas
Computer vision, NLP, reinforcement learning, retrieval-augmented generation, model deployment and monitoring

Projects

HybridRAGRAG · MLOps

A production-style hybrid retrieval architecture combining vector and graph search for more precise LLM grounding.

View repository ↗
PuppyDBPython · Vector search

A custom vector database with LMDB persistence and approximate nearest-neighbor indexing, built to explore high-performance embedding retrieval.

View repository ↗
Neural-ReconComputer vision · Diffusion

An fMRI-to-image reconstruction pipeline using diffusion models and CLIP-aligned features for vision-neuroscience experiments.

View repository ↗
GridWorld-PPOReinforcement learning

PPO-based path planning in dynamic grid environments, including reward shaping and policy evaluation.

View repository ↗
Retrieval-RerankerRAG · NLP

Retrieval and reranking benchmarks across BEIR-style workflows to improve relevance and downstream answer quality.

View repository ↗
More projects
reddit-recsysMultimodal recommendations with CLIP embeddings and vector search.
FinetuningReproducible LLM post-training and evaluation workflows.
exploring-transformersHands-on transformer architecture and tokenizer experiments.
mono-cam-DeepVOMonocular camera motion estimation and trajectory prediction.
BERT sentiment MLOpsFine-tuning, API deployment, monitoring, and automated retraining.
SkinNERClinical named-entity recognition with spaCy and BERT.
Algae ClassificationDeep-learning classification for algae detection and water quality.

Writing

1.58 BitNet ↗

How BitNet-style quantization reduces memory requirements while preserving strong LLM performance.

All posts on Medium ↗

Education & credentials

Indiana University Bloomington2023 — 2025
Master of Science in Data Science · GPA 3.90 / 4.00
Anna University2015 — 2019
Bachelor of Engineering in Computer Science · GPA 8.39 / 10.00
AWS Certified Solutions Architect — AssociateAWS
Certified Kubernetes AdministratorCNCF
Certified Kubernetes Security SpecialistCNCF
Terraform AssociateHashiCorp
NVIDIA-Certified Associate: AI Infrastructure and OperationsNVIDIA

Connect

I’m open to full-time AI/ML and MLOps roles in the United States and available to interview.

GitHub LinkedIn Email Résumé

sarabeshnr@gmail.com · +1 930 333 4115