Generative AI & Agentic ML Engineer

Sonu Kumar

South Delhi, India +91-898******* ************@*****.*** LinkedIn GitHub Kaggle Professional Summary

ML Engineer with an M.Tech in Computer Science (IIIT Delhi), specializing in Generative AI, Agentic AI, and LLM applications. Built production RAG pipelines, NL-to-SQL systems, multi-agent workflows, and MCP servers with Agno, LangGraph, and Amazon Bedrock, deployed on AWS and GCP using FastAPI, Docker, and CI/CD; delivered an 87% accurate NL-to-SQL assistant and a clinical RAG system that cut literature review time by 60%. Education

IIIT Delhi – M.Tech, Computer Science and Engineering (CGPA: 7.15) July 2023 – Apr 2025 Shershah Engineering College, Sasaram – B.Tech, Computer Science and Engineering (CGPA: 7.6) 2019 – 2023 Technical Skills

Generative AI & LLMs: RAG, Agentic Workflows, Prompt Engineering,Fine-Tuning (SFT, PEFT, LoRA, QLoRA), LLMOps, Multimodal Models, NL-to-SQL, Hallucination Detection,Cohere Agentic AI: AI Agents, Tool/Function Calling, Planning, Memory, Multi-Agent Orchestration,LangGraph,Agno, FastMCP NLP & Retrieval: Hugging Face Transformers, BERT, Tokenization, NER, Text Classification, Summarization, Embeddings, Semantic/Hybrid Search, Reranking, Vector Databases Machine Learning: Supervised/Unsupervised Learning, Regression, Classification, Clustering,Random Forest, Scikit-learn, Feature Engineering/Selection, Feature Stores, Pattern Recognition, Model Evaluation, Cross-Validation, Hyperparameter Tuning Deep Learning: PyTorch, CNNs, RNNs, LSTMs, Transformers, Attention Mechanisms, ViT Backend & Databases: Python, FastAPI, REST APIs, Socket.IO, SSE, Next.js, PostgreSQL, SQL, Redis, Neo4j Cloud & MLOps: Model Deployment, Docker, CI/CD, Jenkins, AWS (Bedrock, EC2), GCP, MLOps Languages & Visualization: Python, C, C++, SQL, Plotly, Recharts, Metabase, Power BI Experience

Minds NXT – ML Engineer (Full-Time) May 2025 – Present Agentic Data Analytics Chatbot (NL2SQL) Agno, AWS Bedrock, FastAPI, Socket.IO, PostgreSQL, Recharts

• Built a real-time analytics chatbot replacing fragmented Excel reporting, reducing KPI report turnaround by 70%; NL-to-SQL agent on AWS Bedrock achieved 87% accuracy with minimal hallucination.

• Designed a two-agent pipeline (insights, then statistical analysis and charts) with conversational memory, improving follow-up relevance by 35%.

NeuroPsyGPT: Clinical RAG Query-to-Report Agno, Cohere Rerank, Redis, Qdrant, asyncio, Plotly

• Architected a multi-agent RAG system retrieving evidence concurrently from 15+ biomedical sources, reducing clinical literature review time by 60%.

• Implemented evidence normalization, Cohere Rerank v3.5, and a Citation Gate, improving retrieval relevance by 25% and cutting unverified citations by 80%.

Patient Emotion & Behaviour Detection on IoT Edge ViT, Qwen2-VL, Raspberry Pi 4, AWS EC2, FastAPI, GCP

• Fine-tuned ViT on AWS EC2 for 7-class emotion detection (90% accuracy, 0.3s latency); integrated Qwen2-VL for behavior descriptions from 15 FPS Raspberry Pi 4 video.

• Built a FastAPI server, dashboard, and daily MLOps pipeline to GCP, eliminating manual result handling. Projects

SAP O2C Graph-Based Multi-Agent System LangGraph, Knowledge Graphs, Sentence Transformers

• Built a knowledge graph and LangGraph multi-agent assistant (60+ intents, 92% accuracy); broken-flow analysis cut audit time by 50%.

Detection and Classification of Brain Tumors PyTorch, CNN, Image Augmentation, MRI

• Trained a CNN on MRI scans for tumor detection and classification, achieving 97% accuracy, precision, recall, and F1-score with only 14 false positives and 13 false negatives. Recipe Text Summarization Using LLMs BART, PyTorch, Hugging Face, BeautifulSoup

• Scraped 14,000+ recipes with BeautifulSoup and fine-tuned BART for automated summarization, achieving a ROUGE-1 score of 50.41 (52 on the evaluation set).

Contact this candidate