Open to opportunities

Jigar Purohit

@jigarpurohit

Message

I build low-latency AI, ML, and GenAI systems for banking and industrial automation.

United States

Message

What I'm looking for

I'm looking to build scalable, low-latency AI, ML, and Generative AI systems where strong MLOps, distributed data infrastructure, observability, and risk-aware production engineering matter.

At Citigroup, I engineer deterministic agentic KYC onboarding and payment-fraud systems that reduce compliance review time, control inference costs, and strengthen real-time risk detection.

I build production Generative AI infrastructure with LangGraph, FastAPI, pgvector, LoRA, DeepSpeed, vLLM, TensorRT-LLM, Kubernetes, and GitOps. My work has reduced classifier memory footprints by 35%, increased GPU cache efficiency by 22%, and reduced false-positive transaction alerts by 30%.

Previously at Bosch, I built serverless streaming pipelines processing 4.2 TB of machine-sensor data daily and delivered real-time manufacturing quality and predictive-maintenance models. These systems improved product yield by 1.8%, increased quality prediction accuracy by 2.2%, and reduced unplanned downtime by 29%.

I'm focused on scalable, observable ML systems that operate reliably under strict operational and regulatory constraints.

Experience

Work history, roles, and key accomplishments

Current

AI/ML Engineer

Current

Citigroup Inc.

Jan 2026 - Present (9 months)

Engineered a deterministic agentic KYC onboarding system using LangGraph, FastAPI, and pgvector, reducing manual compliance review times by 20% and cutting token overhead costs by 15%. Executed parameter-efficient fine-tuning with LoRA and DeepSpeed, shrinking infrastructure memory footprints by 35%.

LangGraph fastAPI Pgvector LoRAs DeepSpeed Hugging Face VLLM Kubernetes Kafka PySpark Feast LightGBM GraphSAGE Gitlab CI MLFlow Grafana Evidently AI

Associate Data Scientist

Bosch

Jan 2022 - Jul 2024 (2 years 6 months)

• Developed manufacturing quality prediction models using Python, Scikit-learn, and XGBoost, applying feature engineering and
model evaluation to identify defect patterns and support production decision-making.
• Improved manufacturing quality prediction accuracy by 2.2% by refining production features, tuning XGBoost models, and
validating performance across different equipment.