Susanta Baidya

I’m a Python Machine Learning Engineer, graduated from IIIT Lucknow with an MSc in Artificial Intelligence and Machine Learning. I specialize in designing and deploying scalable AI systems that combine Generative AI, NLP, Computer Vision, and Agentic AI for real-world applications.

My professional experience includes developing enterprise-grade AI analytics platforms and conversational intelligence systems that leverage LangChain, GPT-4, and multi-tenant architectures to automate workflows, enable natural language data querying, and enhance industrial decision-making through predictive analytics.

I’ve also worked on impactful projects such as a lightning prediction system, defect detection using YOLOv5, and RAG-based semantic search engines, integrating deep learning models with cloud and data engineering tools for production-ready deployment.

With a strong foundation in Python, SQL, PyTorch, TensorFlow, FastAPI, and AWS, I focus on building robust, efficient, and explainable AI solutions that bridge data science and business value.

Beyond work, I actively participate in Kaggle competitions, hackathons, and open-source AI projects, continuously exploring how LLMs and autonomous agents can revolutionize automation, analytics, and decision intelligence.

💡 Passionate about applying AI to solve data-driven challenges that make a tangible impact.

Selected Projects

Spacer Defect Detection Link

Advanced computer vision system using YOLOv5 for detecting 23 defect classes, achieving 0.56 mAP@0.5 with Flask-based API and real-time visualization.

Computer VisionYOLOv5FlaskPython

RAG-Based Semantic Quote Retrieval Link

Complete RAG pipeline with semantic search using Sentence Transformers and FAISS vector indexing, processing 2,507+ quotes with 100% query success rate.

RAGFAISSTransformersNLP

CLI Agent - Agentic AI System Link

Advanced agentic AI system with fine-tuned LLM that autonomously converts natural language instructions into precise command-line operations with intelligent planning.

LLMCLIAgentic AINLP

Thunderstorm Prediction System Link

Deep learning-based prediction system using bidirectional LSTM, achieving 69% accuracy with 12-hour lead time for Uttar Pradesh weather forecasting.

LSTMDeep LearningWeather PredictionFlask

Work Experience

Developed optimized pipelines for object detection and defect detection using YOLO, implementing custom algorithms for Adani grading ring object detection that improved detection and inference speed by 35%.
Implemented various detection algorithms for manufacturing defect identification, creating streamlined training workflows that improved accuracy by 28%. and reduced false positives by 45%.

Computer VisionYOLOObject DetectionManufacturing AI

Developed optimized pipelines for object detection and defect detection using YOLO, implementing custom algorithms for Adani grading ring object detection that improved detection and inference speed by 35%.
Implemented various detection algorithms for manufacturing defect identification, creating streamlined training workflows that improved accuracy by 28%. and reduced false positives by 45%.

Computer VisionYOLOObject DetectionManufacturing AI

Developed thunderstorm prediction system for Uttar Pradesh using bidirectional LSTM, achieving 69% accuracy and 0.74 AUC score with a 12 hour lead time by integrating multiple meteorological data sources
Developed Flask-based web interface with real-time lightning tracking and geospatial visualizations, creating risk dashboards and automated ETL pipelines that improved emergency response efficiency by 40%

LSTMDeep LearningWeather ForecastingFlaskGeospatial Analysis