# Subhajit Saha Subhajit Saha is a remote worker on Himalayas, Senior Machine Learning Engineer at Quantiphi, based in India. Source: https://himalayas.app/@subhajitsaha ```text Senior AI/ML Engineer | Generative AI & Agentic AI | LLM, RAG, AWS, LangGraph, Fine-Tuning (RLHF), Multi-Agent Systems | Evaluation | Observability ``` | Detail | Value | | --- | --- | | Availability | Looking for a job | | Location | India | | Authorized to work in | India | | Salary expectations | 1800k-2400k INR | | Job categories | AI Engineer, AI Engineer, Generative AI Engineer, Agentic Systems Engineer | | Skills | Machine Learning, Agentic AI, Multi Agent Systems, Generative AI, RAG, Agentic RAG, LLM orchestration, LangGraph, Lang Chain, AWS, Python, LLM Evaluation, Prompt Engineering, GuardRails, LLM Observability, NLP, Computer Vision, Multi Agent Frameworks, Multimodal AI, SQL, Docker, LLM Fine tuning | | LinkedIn | https://https://www.linkedin.com/in/subhajit-saha-118135198/ | ## About ```text I'm a Senior AI/ML Engineer at Quantiphi with 4+ years of experience building and shipping production-grade AI systems, currently focused on agentic AI and Generative AI applications on AWS. At Quantiphi, I've worked across the full lifecycle of AI systems — from architecture to production deployment to measuring real business impact. A few things I'm proud of: - I architected an LLM-powered document intelligence pipeline on AWS (Lambda, Textract, Bedrock Claude, S3, DynamoDB) that extracts structured data from multilingual student transcripts, cutting turnaround time by over 80% and eliminating $200 in per-student evaluation cost. - I designed a multi-agent LLM orchestration workflow using LangGraph — with parallel execution and tool-calling — to extract and validate entities from semi-structured invoices, reducing manual review effort by roughly 60%. - I shipped an end-to-end RAG system for a pharma client and built an RLHF fine-tuning pipeline for LLaMA 3.1 using PEFT/LoRA and quantization. - I have implemented an LLM-as-a-judge evaluation layer to automatically score and monitor production LLM outputs — because I think evaluation and observability are as important as the model itself once something is in production. - I've also worked on regulated, compliance-sensitive systems: a dental claims extraction pipeline (LayoutLM + YOLOv8) that reduced processing time by 70%, with automated PII redaction built into the pipeline. Outside of client work, I build agentic and multimodal systems on my own time — a multi-hop Agentic RAG chatbot combining FAISS and GraphRAG for knowledge-graph retrieval, extended to handle audio and video via Whisper, AWS Transcribe, and CLIP embeddings, and a CrewAI-based multi-agent system for automated data analysis. My technical interests center on agentic AI systems — how agents reason, plan, use tools, and maintain memory over long-running tasks — along with LLM evaluation and observability, and RAG/retrieval architecture. I hold AWS Certified Solutions Architect – Associate and AWS Certified Machine Learning – Specialty certifications. I'm looking to move into roles with deeper ownership of agent architecture — systems where I'm making the core design decisions on reasoning, memory, and reliability, rather than primarily building on top of existing frameworks — ideally in high-stakes or regulated domains where explainability and auditability matter. ``` ## Experience ### Senior Machine Learning Engineer (current) [Quantiphi](https://himalayas.app/companies/quantiphi-com). Apr 2025 - Present ```text • Architected an LLM-powered document intelligence pipeline on AWS, reducing turnaround time by over 80% and eliminating $200 per-student evaluation cost. Implemented an LLM-as-a-judge evaluation layer. • Engineered a production RAG chatbot on AWS Bedrock using OpenSearch with reranking. Built an RLHF pipeline to fine-tune LLaMA 3.1 using PEFT (LoRA). ``` Skills: AWS, Lambda, Textract, Bedrock, Claude, S3, DynamoDB, RAG, OpenSearch, Reranking, RLHF, PEFT, LoRAs, Quantization, LLM-As-A-Judge, Python, LangGraph, LLM-Fine-tuning ### Machine Learning Engineer [Quantiphi](https://himalayas.app/companies/quantiphi-com). Aug 2022 - Mar 2025 ```text • Designed and deployed a multi-agent LLM orchestration workflow on AWS Bedrock with LangGraph to extract, validate key entities from semi-structured invoices, reducing manual review effort by approximately 60%. • Architected an automated extraction pipeline combining LayoutLM and YOLOv8 to detect PII, checkboxes, and signatures on dental claim forms, driving a 70% reduction in processing time. ``` Skills: AWS, Bedrock, LangGraph, Multi-Agent-Systems, Tool-calling, PII-Masking, LayoutLM, YOLOv8, Model-Retraining, Python, Computer-Vision, NLP ## Education ### Indian Institute of Technology Kharagpur M.Tech, Farm Machinery and Power. 2020 - 2022. Grade: 8.92 CGPA ```text Pursued M.Tech in Farm Machinery and Power, achieving a CGPA of 8.92. ``` ### Bidhan Chandra Krishi Viswavidyalaya B.Tech, Agricultural Engineering. 2016 - 2020. Grade: 8.4 CGPA ```text Completed B.Tech in Agricultural Engineering with a CGPA of 8.4. ``` ## Tech stack - Amazon S3 - NumPy - Pandas - OpenCV - Python - TensorFlow - PyTorch - scikit-learn - Keras - Streamlit - NLTK - FastAPI - Transformers - Amazon DynamoDB - AWS Lambda - Git - Docker - GuardRails - SQL - LangChain - CrewAI - DeepEval - Faiss - LangGraph - LangSmith - PEFT - GraphRAG - AWS To contact Subhajit, message them from this profile with a Himalayas employer account: [Hire on Himalayas](https://himalayas.app/hire). ## People also viewed - [Chandana J](https://himalayas.app/@chandanaj): Agentic Legal Analysis, Courtroom Insight - [Shivam Gulati](https://himalayas.app/@shivamgulati): Data Scientist, Quara AI Tech Pvt Ltd - [abhishek kumar](https://himalayas.app/@abhishekkumar63): Senior GenAI Engineer, Syngene International Ltd - [chejarla chaitu](https://himalayas.app/@chejarlachaitu): Senior Machine Learning Engineer, Motivity Labs - [Sumit Kumar](https://himalayas.app/@sumitkumar34): Machine Learning Engineer IV, Avalara - [ajay ai](https://himalayas.app/@ajayai): Senior AI Engineer, Sutherland - [surajit dey](https://himalayas.app/@surajitdey): Senior Data Scientist, Foundation AI - [Jaydatta Bade](https://himalayas.app/@jaydattabade): AI Engineer Individual Contributor, FilersKeepers/Lawstronaut - [Jyotiranjan panda](https://himalayas.app/@jyotiranjanpanda): Director, GenAI Solution Architect, Firstsource Solutions Ltd - [Prabhat gupta](https://himalayas.app/@prabhatgupta3): ML Engineer, London Stock Exchange Group (LSEG)