Senior ML Engineer specializing in large-scale inference platforms, foundation-model training frameworks, and MLOps infrastructure for generative AI. Currently building ML platform at Adobe Firefly for generative AI workloads over billions of images and videos. Published researcher in Graph Neural Networks and bias mitigation.

Education 🎓

• MS in Computational Data Science — Indiana University, Bloomington (2023) • GPA: 3.87/4.0
• BS in Computer Science — National Institute of Technology, Patna (2015) • GPA: 8.32/10.0

Current Work 💼

• Senior Machine Learning Engineer (ML Platform & Frameworks) @ Adobe Firefly (Jul 2024 - Present)
  • Designed and led the org's online + offline inference systems — vLLM + Ray for LLMs (Falcon-40B, Llama 2 70B); Ray Data + PyTorch Lightning offline inference hitting 3–8K images/sec on 32 GPUs (10–50× faster image loading).
  • Leading development of a PyTorch foundation-model training framework adopted across the org — plugin-based FSDP/FSDP2 strategies, Flash Attention 2/3, and context/distributed attention for DiT text-to-image and text-to-video training.
  • Co-built the underlying Kubernetes platform: plugin pipelines (KEDA-autoscaled on SQS), BuildKit/ECR image builds, a Rust control plane, and an MCP interface so AI agents can drive build/deploy/inference end-to-end.
  • Built the org's first data quality framework on Cerberus, used across enrichment pipelines to validate the correctness of feature-generation outputs.

Previous Experience

• Senior Software Engineer (Data Platform) @ EvolutionIQ (2023-2024) — Data and model-training pipelines at the intersection of generative AI, fintech & healthcare; fairness/bias evaluation framework.
• Software Dev Engineer II (ML Platform) @ Swiggy (2019-2021) — Online inference platform serving 18M+ MTU; feature store (4Bn rows, 10K QPS); founding member of forecasting & correlation platform; DAQ collecting 15M rows/day.
• Software Dev Engineer (Search Relevance) @ Flipkart (2017-2019) — Search intent models (CRF/NN); Flipkart's first automated ML training & deployment workflow (Luigi → Airflow); Hackday 9 Winner.
• Software Engineer @ Groupon, NetSpeed Systems (2015-2017) — Backend engineering & Network-on-Chip modules (Polarity-based Arbitration, Multi-Cast Filtering).

Research 📄

• First Author — "Biased Contrastive Learning debiases Graph Neural Networks" (NetSci 2023, IC2S2 2023)

Inference & Serving: vLLM • Ray • Ray Data • Ray Serve • PyTorch Lightning • TensorRT-LLM • Triton Inference Server

Training Frameworks: PyTorch • FSDP / FSDP2 • Flash Attention 2/3 • Diffusion Transformers (DiT) • Context / Distributed Attention

ML/AI: PyTorch Geometric • TensorFlow • Scikit-learn • LLMs • GNNs • NLP • Computer Vision

Data Engineering: Apache Spark • Apache Kafka • Apache Flink • Hadoop HDFS • Airflow • Feature Stores • Cerberus

Platform & Infra: Kubernetes • KEDA • BuildKit • Docker • AWS (S3, SQS, ECR, SageMaker, DynamoDB) • Model Context Protocol (MCP)

Languages: Python • Rust • Scala • C++ • SQL • Java

🌐 Portfolio: thunderock.github.io