WILLIAM TOM JACOB
+91-747******* # ************@*****.*** ï william-tom-jacob williamtom3010.github.io * Delhi, India PROFESSIONAL SUMMARY
AI Engineer with 2 years of hands-on experience in building and deploying AI solutions. Experienced in RAG-based chatbots, LLM fine-tuning and CPU/GPU deployment, Python and FastAPI backend development, LLM API integration, vector databases, model optimization, and containerized deployments on AWS and Paperspace. Strong foundation in machine learning and deep learning.
TECHNICAL SKILLS
Languages: Python, R, HTML/CSS
Technologies: Generative AI, Large Language Models (LLMs), Retrieval-Augmented Generation (RAG), Agentic AI, Prompt Engineering, Neural Networks, Deep Learning, Machine Learning. Frameworks: Transformers, NLTK, TensorFlow, Numpy, Pandas, Seaborn, Matplotlib, Scikit-learn, Beautiful Soup, OpenCV, PyTorch, FastAPI
Visualization Tools: Tableau
OS Known: Windows, Linux
Databases: PostgreSQL, NoSQL, MongoDB, FAISS DB
API Models Used: OpenAI, Claude, Groq, Gemini
EXPERIENCE
Techvantage AI Apr 2025 – Present
Associate Software Engineer - AI Trivandrum, Kerala
• Developed backend Python services for the ENOL mobile application, implementing damage analysis, and vehicle comparison features using multiple LLM APIs . Built and integrated FastAPI endpoints to connect the Python backend with the React frontend. Designed input schemas tailored to frontend requirements, and implemented MongoDB-based request/response logging .
• Developed and fine-tuned a ResNet-50 transfer learning model for client-side classification of table images into Benefits, Pricing, and Other categories. Evaluated model performance using precision, recall, F1-score, and confusion matrix to identify misclassifications, then retrained the model to achieve approximately 80% precision and 70% recall.
• Built an OCR node for ZENTIS 2.0, a custom Agentic AI framework for the BFSI industry, with support for AWS Textract, Azure AI Document Intelligence, and self-hosted DeepSeek OCR on AWS EC2 instance servers.
• Reduced a complex, multi-step Hugging Face fine-tuning workflow to a few clicks by collaborating with the front-end team to build a UI automating model/dataset loading, fine-tuning, and multi-precision GGUF conversion for Hugging Face deployment.
• Built a speech-to-speech conversational AI pipeline for Indian languages by integrating Speech-to-Text (STT), Large Language Models (LLMs), and Text-to-Speech (TTS) systems to generate context-aware spoken responses.
• Containerized and deployed LLaMA-3.1-8B GGUF and safetensor models for ZENTIS 1.0 on AWS EC2 using vLLM, reducing operational costs and minimizing reliance on external API providers such as OpenAI, Claude.
• Performed rigorous data preparation and cleaning on diverse datasets, ensuring high-quality inputs for fine-tuning and model training pipelines across multiple AI projects. Trainee Software Engineer - AI Jan 2025 – Mar 2025
• Applied statistical modeling to evaluate LLaMA-3.1-8B fine-tuning performance; optimized for CPU inference via llama-cpp (GGUF), achieving a <2% performance drop compared to GPU across Semantic, ROUGE-L, and BLEU metrics, validated by GPT-4.1 automated evaluation.
• Performed CPT on Mistral-7B with actuarial domain corpus followed by SFT. Built domain-specific QA dataset and evaluated improvements on ARC and MMLU benchmarks.
• Performed in-depth data analysis on HR datasets to curate a custom QA dataset, applying synonym-based augmentation and L1-Norm pruning on Mistral-7B to optimize model footprint while maintaining accuracy.
• Developed a context-aware chatbot using Streamlit that automatically routed user queries to the most suitable LLM. Stored queries and responses in MongoDB and PostgreSQL relational databases, leveraging SQL for efficient data extraction and analysis.
Aavaaz -Internship Dec 2024 – Jan 2025
AI Engineer Intern Chicago, Illinois, United States
• Built an NLP-based emotion classification model to classify English Twitter messages into six emotion categories: anger, fear, joy, love, sadness, and surprise.
• Performed feature engineering and text preprocessing (stopword, punctuation, non-alphabetic tokens removal), to reduce textual noise. Applied CountVectorizer for Bag-of-Words feature extraction.
• Trained Logistic Regression model and Achieved 89% accuracy and assessed classification performance using precision, recall, F1-score, and confusion matrix. Suvidha Foundation (Suvidha Mahila Mandal) -Internship Nov 2024 – Dec 2024 Machine Learning Intern Nagpur, Maharashtra
• Collected and preprocessed ( stopword and punctuation removal) a dataset of 1000 Indian news articles and used to train text summarization task. The transformer model used for training is BART. Light Logics Holography and Optics Pvt. Ltd. -Internship May 2023 – Jun 2023 Machine Learning Intern Trivandrum, Kerala
• Developed a service in detecting the contour and identifying the the dark and bright fringes using OpenCVpackage.
• Gained comprehensive skills in both basic and advanced OpenCV techniques. PROJECTS
RAG-Based Intelligent Chatbot for Vehicle Manual Assistance Python, Deep Learning November 2025
• Developed a Retrieval-Augmented Generation (RAG) chatbot using FAISS for vector storage and embeddings of the Suzuki Fronx car manual, enabling accurate context-aware responses through semantic document retrieval. Semantic segmentation on iris(eye) dataset Python, Deep Learning July 2024 – August 2024
• Developed a Computer Vision solution for semantic segmentation using a U-Net architecture (Convolutional Neural Network) on an iris dataset, achieving precise mask generation for 250 subjects. Student Centered automated feedback system using transformer(M Tech Dissertation) Oct 2023 – June 2024
• Designed a automated feedback system for students using the transformer model GPT-2 to train on Indian History dataset. Conducted data augmentation to enhance dataset size and quality. Predictive maintenance using machine learning approach (Mtech mini project) April 2023 – Aug 2023
• Implemented predictive tools for maintenance scheduling using machine learning models. The machine learning algorithm used for training was Support Vector Machine.
• A Synthetic data set which is obtained from Kaggle is used for predictive maintenance. Utilized SMOTE to address data imbalance issues.
EDUCATION
Master of Technology in Artificial Intelligence Oct. 2022 – June 2024 APJ Abdul Kalam Technological university CGPA : 9.13/10.0 Bachelor of Technology in Mechanical Engineering 2017 – 2021 National Institute of Technology,Bhopal CGPA : 7.06/10.0 PROFESSIONAL TRAINING
Crew AI Certificate Course May 2025
IBM Data Science Professional Certificate Course Jan 2025 – Feb 2025 Internshala Training (Data Science Certification Course) March 2024 – June 2024 Natural Language Processing-IIT Kharagpur (NPTEL) Oct 2023 Data Analytics with Python-IIT Roorkee (NPTEL) Apr 2023