Create your free profile

About

AI/ML Engineer with 1.8 years of experience at TCS building applied machine learning and Generative AI solutions for automotive engineering teams. Built a Graph Neural Network surrogate that predicts full restraint-force curves from LS-DYNA simulations (8 hours to 2 minutes), a stiffness-prediction model integrated into an engineering GUI, and a RAG-based engineering assistant. Strong in PyTorch, PyTorch Geometric, scikit-learn, model evaluation, LLMs/RAG, and Python engineering workflows.

Experience1

  1. AI/ML Engineer

    Jan 2025 — Present

    Tata Consultancy Services (TCS)

    Designed a Graph Neural Network surrogate in PyTorch Geometric reducing LS-DYNA simulation time from 8 hours to 2 minutes by predicting full restraint-force curves. Parsed 1,030 LS-DYNA files to create graph datasets used for training/validation/testing. Developed stiffness prediction models using various regressors achieving 22.74 N/mm test MAE and R²=0.896, integrated into Qt engineering GUI. Created a RAG-based engineering assistant reducing documentation search times from ~5 hours to 3 minutes. Automated engineering data validation and reporting, improving task durations from days to minutes with Python and Streamlit.

Projects2

  • Domain-Specific QA RAG

    Built a RAG pipeline over 1,500+ documents with hybrid BM25 + dense retrieval and cross-encoder reranking, improving Recall@5 from 68% to 84% and MRR from 0.55 to 0.71 on 100 labeled questions. Fine-tuned Mistral 7B with LoRA/QLoRA (4-bit), enhancing faithfulness by 22% and halving hallucination rate. Reduced end-to-end latency from 4.1s to 1.8s with caching, batching, quantization; containerized with Docker and served via FastAPI and Streamlit interface.

    • Python
    • Hugging Face
    • FAISS
    • LangChain
    • LoRA
    • QLoRA
    • FastAPI
    • Docker

  • User Conversion Prediction

    Built conversion prediction models on 12,330 e-commerce sessions with 15.5% conversion rate handling class imbalance. Achieved ROC-AUC 0.928 and PR-AUC 0.742, identifying 79.8% of converters to enable targeted marketing strategies.

    • Python
    • scikit-learn
    • XGBoost
    • pandas

Education1

  • B.Tech

    2020 — 2024

    National Institute of Technology, Warangal

    Mechanical Engineering