🎯 AI/ML Engineer | LLM & Generative AI Specialist
🔍 Building real-world solutions with Deep Learning, NLP, and Computer Vision

🌐 LinkedIn | ✍️ Medium | 📧 er.rajkumaar@gmail.com

AI/ML Engineer with 14+ years of experience in banking, analytics, and enterprise-grade automation. I’ve spent the last 6+ years solving real-world problems using deep learning, LLMs, and computer vision — from smart document QA and chatbot systems to secure image generation and anomaly detection.

Specialized in building explainable, ethical, and inclusive AI systems with a strong focus on NLP for low-resource Indian languages and scalable deployment using open-source tools.

• 🤖 LLMs & Generative AI

  • End-to-end chatbot and document QA system development using Hugging Face Transformers, LangChain, and custom LLMs
  • Retrieval-Augmented Generation (RAG) pipelines using FAISS, Pinecone, and embedding stores
  • Prompt engineering for zero-shot, instruction-based, and multilingual LLM applications
  • Low-resource language model development (e.g., Angika GPT) for regional translation and content generation
  • Scalable deployment of LLM APIs with FastAPI and Flask

• 📚 Natural Language Processing (NLP)

  • NER, Sentiment Analysis, Semantic Search, and Text Classification using transformer-based architectures
  • Translation systems for multilingual and low-resource Indian languages
  • Fine-tuned models for summarization, question answering, and domain-specific content generation
  • Custom NLP pipelines built using Hugging Face, spaCy, and NLTK

• 📈 Machine Learning & Forecasting

  • Forecasting and risk modeling using ARIMA, XGBoost, and LSTM
  • Ensemble learning, advanced feature engineering, and hyperparameter optimization
  • Model interpretability and fairness via Explainable AI techniques (SHAP, LIME)
  • Full lifecycle ML pipeline implementation with scikit-learn and PyTorch

• 🎥 Computer Vision & Generative AI

  • Object detection, segmentation, OCR, and surveillance-based anomaly detection using YOLO, SAM, SegFormer, and OpenCV
  • Built generative pipelines for Text-to-Image, Image-to-Image, and Image-to-Video using Diffusion Models, DCGANs, and ControlNet
  • Face image encryption and security through GAN-based adversarial training
  • Applied and experimented with Large Vision Models (LVMs) like Florence and SegFormer

• ⚙️ MLOps & Deployment

  • Designed and deployed modular ML APIs using Flask, FastAPI, and Streamlit
  • Model packaging with Pickle, Hugging Face Hub, and CI/CD pipelines via GitHub and GitLab
  • Experience with cloud-agnostic and secure on-prem enterprise deployments
  • Version control, reproducibility, and lifecycle tracking using Git-based workflows

• 📊 Analytics & Data Engineering

  • Data preprocessing, transformation, and integration from SQL, MySQL, and DB2
  • Advanced analytics including regression, hypothesis testing, and A/B testing
  • Visualization using Matplotlib, Seaborn, Plotly, and Power BI for decision support

• 👥 Leadership & Agile Delivery

  • Agile/Scrum-based project management with JIRA and Confluence
  • Sprint planning, roadmap building, stakeholder communication
  • Led globally distributed teams across delivery lifecycles from PoC to production
  • Mentored engineers on Python, ML, NLP, and deployment standards
  • Translating business goals into scalable and interpretable AI solutions

• Certified SAFe® Scrum Master
  Issued by Scaled Agile, Inc. | Issued March 2025
  Credential ID: 48725598-4060

Delivered enterprise-grade AI/ML solutions across banking, finance, and language tech using LLMs, Computer Vision, and Deep Learning.

Flask + LangChain powered RAG chatbot with multi-doc upload, image-aware retrieval, session memory, and a clean Markdown chat UI using ChromaDB and Gemini.

A computer vision-based alert system to detect heat spikes in thermal or surveillance videos using OpenCV. Designed for industrial safety and anomaly detection with real-time alert logging and visualization.

A modular computer vision pipeline that reads driving scene videos and overlays category-specific annotations (bounding boxes and labels) from CSV files. Built using OpenCV and Python for rapid prototyping, visual validation, and AI model support in video analytics workflows.

A computer vision project that detects vehicles, estimates speed, flags over-speeding, and generates alert frames — built with YOLOv8, OpenCV, and Jupyter.

Developed a personalized ML-based lending solution for early-stage retailers with 2–3 years of business history. Utilized multiple models to evaluate creditworthiness and integrated Explainable AI (XAI) for transparent, data-driven decision-making. Improved loan approval accuracy while reducing default risks.

Developed a neural machine translation (NMT) model to translate Angika—an endangered Indian language—into English. Used an encoder-decoder LSTM architecture in TensorFlow Keras. Aimed at preserving linguistic heritage and enabling accessibility for low-resource languages. Supports cultural conservation through AI-driven translation.

Leveraging my expertise in Natural Language Processing and Generative AI, I spearheaded the development of Angika GPT, (www.angikaGPT.com) a custom Language Model designed to address the challenges of low-resource language translation and content generation, with a particular focus on Angika. This initiative involved utilizing Hugging Face Transformers to build and fine-tune a specialized LLM, actively contributing to digital preservation efforts by enabling content creation and translation in Angika. My role encompassed all aspects of the project, from model selection and training to performance evaluation and optimization, ultimately fostering the development of inclusive language technology and promoting multilingual NLP capabilities.

ML model predicting story points, impacted components, and risk levels using historical JIRA data and NLP — improved sprint planning by 20%.

Real-time assistant using LangChain + RAG for invoice-related queries. Integrated document loaders, chunking, embeddings, and vector search for fast, human-like responses. Included intent detection and adaptive learning.

Improved invoice/cheque OCR pipeline using DL for handwritten and printed text — reduced manual processing by 25%.

Built a predictive alert dashboard with real-time updates, anomaly detection, and visual trends — cut incident response time by 30%.

• 🧠 Explainable Sentiment Analysis for Financial News – ICIEM 2023
• 🌐 NMT for Endangered Languages (Angika) – IIT Patna, 2023
• 💊 Generative AI for Drug Recommendations – SHADG 2024 (IIT Kharagpur)
• 📊 LLMs & DPT Models for Sentiment Analysis – NIT Patna 2024
• 🔐 Facial Image Encryption with DCGANs – M.Tech Project, IIT Jodhpur 2023

• Led cross-functional AI/ML teams across geographies for complex delivery programs
• Acted as Agile Scrum Master — driving velocity, alignment, and outcomes
• Mentored early-career engineers in Python, ML, NLP, and deployment workflows
• Modernized legacy platforms by embedding AI/ML pipelines
• Defined roadmaps and aligned data science goals to business strategies

“AI is not just about intelligence — it's about accessibility, ethics, and trust-driven innovation.”
— Raj Kumar