Shubham Goel

AI / Generative AI Engineer | Delhi, India

About Me

AI and Generative AI Engineer with a Master's in Applied Mathematics from NYU, passionate about building intelligent systems that solve real-world challenges. With strong expertise in Python, deep learning frameworks like PyTorch and TensorFlow, and tools such as LangChain, I specialize in fine-tuning LLMs, developing multi-agent pipelines, RAG systems, and automation solutions — from GUI agents and legal AI to sentiment analysis and recommendation engines. Committed to leveraging mathematics, machine learning, and generative technologies to create efficient, scalable, and impactful AI applications.

Shubham Goel

Skills

Python
C++
PyTorch
TensorFlow
LangChain / LangGraph
LLM Fine-Tuning (LoRA)
Retrieval-Augmented Generation (RAG)
Hybrid Search (Dense + BM25)
InLegalBERT / Custom Embeddings
Computer Vision
Natural Language Processing
Generative AI
Multi-Agent Systems
ChromaDB
Kubernetes
Git
Project Management

Education

New York University, NY, USA

Master of Science, Applied Mathematics
Sep 2023 – May 2025

Guru Gobind Singh Indraprastha University, New Delhi, India

Bachelor of Technology, Electronics and Communications Engineering
Sep 2008 – May 2012

Work Experience

Development Intern (ML), Insurance Samadhan, Noida, India

Jun 2024 – Aug 2024

  • Implemented chatbot system for extracting user-specific details from insurance policy documents; designed end-to-end pipeline processing 2.3M+ data points; fine-tuned DistilBERT to 95% accuracy, improving retrieval efficiency by 20%.
  • Architected TensorFlow-based image processing pipeline using U-Net to remove watermarks; processed 1.2M+ images, reducing data extraction time by 30%.

Projects

Legal AI Copilot

• Built an AI Legal Copilot using FastAPI, Next.js, and a multi-agent RAG pipeline to process legal queries and retrieve case laws (~80K+ scalable architecture); reduced manual research time by ~70% via automated retrieval and drafting. • Designed a modular multi-agent system (Reasoning, Outcome, Judge, Drafting) with human-in-the-loop refinement; improved answer reliability by ~40% using adversarial judge feedback to identify gaps and strengthen arguments. • Developed a full-stack app with FastAPI backend (ngrok) and Next.js (Tailwind) frontend on Vercel; enabled real-time interaction with multiple output modes, achieving ~5–10s response latency per query.

View on GitHub →

AI Knowledge Assistant

• Built an AI-powered knowledge assistant to extract answers from large documents, instead of manually searching through **hundreds of pages**. • Developed a pipeline that ingests **3 full-length books (~330K characters)**, processes them into **hundreds of searchable text chunks**, and retrieves the **top-3 most relevant passages** for every user query.• Delivered a **real-time document question-answering system** that generates responses grounded in retrieved context, enabling faster information discovery and reducing incorrect or hallucinated answers.

View on GitHub →

VLM-based GUI Automation on Cloud Environments

VLM-driven pipeline: 83% success on 236 Google Cloud workflows (≤20 prompts). Created 10k triple dataset. Multi-agent system (planner + executor LLM) with persistent browser.

View on GitHub →

Time Series Analysis of NYC Temperature Data

Analyzed 528 monthly records (1970–2013). EDA, seasonal decomposition (0.45°C/decade rise). Compared ARIMA/SARIMA/Prophet; SARIMA selected (RMSE 1.55°C) for 20-year forecasts.

View on GitHub →

Movie Recommender System

Used MovieLens + TMDB. SVD (RMSE ~0.85) & TF-IDF models (50% Precision@10). Interactive Plotly visuals + deployed Streamlit web app for instant recommendations.

View on GitHub →

Mental Health Sentiment Analysis

Processed 53k+ social media posts → 20k unique. VADER sentiment (66% accuracy), TF-IDF + LDA topic modeling (5 topics e.g. Depression 23%), chi-square validation. Visualizations: word clouds, heatmaps for insights.

View on GitHub →

COVID-19 Data Analysis & Visualization (Jun 2025)

Developed 10+ advanced SQL queries (joins, CTEs, subqueries) on Our World in Data. Built interactive Tableau dashboards (maps, heatmaps, scatter plots) showing global trends, case growth, policy impacts, and correlations. Hosted on GitHub and Tableau Public.

View on GitHub →