AI Engineer with 3+ years of experience building proprietary middleware and agentic workflows to drive operational efficiency. Specialized in Generative AI fine-tuning (RLHF, GRPO, LoRA) and deploying high-throughput, cloud-native solutions on GCP/Vertex AI and AWS.

• AI & Agentic Systems: LangGraph (State Graphs), CrewAI, Prompt Engineering, RAG Pipelines (LangChain), Agentic Feedback Loops, GitHub Copilot
• Fine-Tuning & Eval: RLHF, GRPO, LoRA/QLoRA, SFT, LLM Evaluation (BLEU, ROUGE)
• Engineering & Ops: FastAPI, SQL, Streamlit, React, Docker, Git, GitHub Actions
• Clinical & Compliance: HIPAA, GDPR/ENISA, RBAC, FHIR Standards, DICOM
• Cloud & Hardware: GCP (Vertex AI), AWS, NVIDIA H100, AMD MI300X, vLLM, PySpark

• Built a multimodal assistant using MedGemma 4B for real-time medical image analysis (X-ray, MRI) and SOAP note generation.
• Architected a HIPAA-compliant RAG pipeline for patient context retrieval using FHIR-compliant EHR data and encrypted DICOM images.
• Developed a LangGraph-based state graph where multiple MedGemma 4B instances execute in parallel to compute independent differential diagnoses and inter-agent agreement.

• Developed a fine-tuning framework at the intersection of SFT and GRPO using LoRA, Unsloth, and Hugging Face.
• Optimized model inference using vLLM's PagedAttention and multi-GPU setups to ensure low-latency, real-time reasoning.

• Developed a LangGraph workflow utilizing Chain-of-Thought (CoT) patterns and a RAG architecture to analyze complex soccer queries.
• Used vLLM's PagedAttention and multi-GPU setups to ensure low-latency responses for real-time agentic reasoning.

• Deployed a high-concurrency AlphaZero agent on Hugging Face Spaces.
• Engineered the communication infrastructure to handle real-time state updates.
• Optimized model inference for a web-based environment to serve complex AI models via serverless-style hosting.

• Orchestrated a PySpark and BigQuery pipeline to process over 1 million images for real-time feature extraction.
• Developed models for object detection, showcasing the ability to handle massive-scale image data pipelines.

Freelance | Machine Learning Specialist

• Deployed a custom "Morning Briefing" multi-agent system using CrewAI to automate patient data synthesis and streamline clinical reporting.
• Executed RLHF and GRPO fine-tuning on NVIDIA H100 and AMD MI300X GPUs to optimize model performance for clinical safety.

Sopra Steria | Machine Learning Engineer

• Engineered a RAG pipeline using LangChain and ChromaDB to transform static GDPR/ENISA policy documents into interactive agentic interfaces.
• Developed FastAPI-based middleware and BiLSTM models to automate the routing of 10,000+ monthly service tickets, reducing manual triage by 40%.

• MS in Applied Data Analytics, Boston University (Jan 2024 - May 2025).
• Research Assistant (Reinforcement Learning): Architected an open-source C++ package integrating LibTorch with OpenAI Gym for high-performance RL including A2C and PPO.
• Teaching Assistant (Big Data Analytics): Led sessions on deploying PySpark solutions across GCP and AWS for large-scale batch and streaming.

• Email: 23.deepak.s@gmail.com
• Links: LinkedIn | GitHub | AI Writeups