Machine Learning · NLP · Financial AI · Computer Vision · Software Engineering

Computer Science Engineering undergraduate building reliable machine learning systems, multi-agent architectures, and production-oriented software.

I am a Computer Science Engineering undergraduate specializing in machine learning, natural language processing, financial AI, and agentic workflows.

My focus is on engineering full-lifecycle AI systems: spanning multimodal data ingestion, high-dimensional vector search, econometrics and causal inference, backend orchestration with FastAPI/Celery, and production deployment.

I actively contribute to major open-source repositories. Notably, I contributed an accessibility fix to the Microsoft Visual Studio Code codebase correcting keyboard shortcut interpretations for screen readers, which was merged into upstream VS Code.

Autonomous Multi-Agent Enterprise Catalog & Product Intelligence Platform

• An autonomous multi-agent cognitive architecture (Web Research, Multimodal Vision, Catalog RAG, and LOV Guardrails) powered by Google Gemini 2.5 Flash and Next.js 15.
• Transforms unstructured manufacturer part numbers and technical PDF specification datasheets into validated 252-column enterprise commerce catalogs with automated confidence scoring.
• Tech: TypeScript, Next.js, React, Google Gemini 2.5 Flash, Tailwind CSS, Vercel Edge.
• 🌐 Live Demo

Real-Time UPI Fraud Detection & Causal Inference Platform

• A production-grade financial fraud intelligence system combining a 4-tier consensus ensemble (Z-score, Isolation Forest, RBI heuristics, and Meta FAISS L2 vector similarity search over 29,000+ fraud signatures).
• Quantified the causal impact of security interventions using Difference-in-Differences (DiD) econometrics, proving that mandatory 2FA reduces fraud probability by 0.73 percentage points ($p = 0.013$).
• Built on a 3-container Docker stack (FastAPI + Celery + Redis) handling 1,000,000+ calibrated transactions with 29 automated test suites.
• Tech: Python, FastAPI, Celery, Redis, Docker, Meta FAISS, Scikit-learn, Statsmodels, Power BI, Streamlit.

Bilingual Medication Safety & Interaction Knowledge Graph Engine

• A real-time medication verification platform designed for the Indian healthcare ecosystem, featuring multi-strategy OpenCV preprocessing and bilingual EasyOCR (English + Devanagari) for blister pack text extraction.
• Models drug contraindications as a weighted NetworkX knowledge graph across 65 essential formulations and 55+ interaction clusters, coupled with a personalized polypharmacy risk model adjusting for patient age and organ impairments.
• Features a dark medical-grade PWA with live camera auto-scanning over HTTPS and interactive Canvas graph visualization.
• Tech: Python, FastAPI, PyTorch, EasyOCR, OpenCV, NetworkX, RapidFuzz.

AI-Powered SEC 10-K Financial Filing RAG Analyzer

• An agentic RAG analysis platform that ingests complex SEC 10-K financial filings, parses tables and disclosures, and answers natural language investor questions with sentence-level citations.
• Orchestrates retrieval using LangGraph, Meta FAISS vector index, and Groq Llama 3 for low-latency reasoning.
• Tech: Python, LangGraph, Groq, FAISS, Streamlit, NLP.
• 🌐 Live Demo

Real-Time Road Accident Detection & Automated Emergency Dispatch

• A computer vision and event-driven pipeline that identifies traffic collisions from live surveillance feeds using OpenCV and deep learning.
• Automatically triggers GPS-calibrated emergency routing and automated ambulance dispatch to minimize emergency response latency.
• Tech: Java, OpenCV, Computer Vision, Deep Learning.
• 🌐 Live Demo

Interactive Interior Design & Spatial Optimization Visualizer

• An AI-assisted interior planning tool leveraging Gemini Multimodal Vision to generate room layouts and spatial enhancements.
• Ranked as the most popular project on profile (⭐ 5 Stars, 4 Forks).
• Tech: TypeScript, Gemini Vision, React, Canvas API.
• 🌐 Live Demo

I actively contribute to established open-source repositories spanning developer tooling, bio-informatics, and accessibility:

• Languages: Python, TypeScript, JavaScript, Java, C++, SQL, HTML/CSS
• ML & AI Frameworks: PyTorch, TensorFlow, Scikit-learn, XGBoost, Meta FAISS, Sentence-Transformers, LangGraph, EasyOCR, OpenCV
• Backend & Cloud: FastAPI, Next.js, Celery, Redis, Docker, PostgreSQL, SQLite, Vercel
• Data & Analytics: Pandas, NumPy, Statsmodels, Scipy, NetworkX, Power BI, Streamlit
• DevOps & Tooling: Git, GitHub Actions, Linux, VS Code, Swagger/OpenAPI

• Delegate at the Harvard Project for Asian and International Relations (HPAIR)
• Participant in the Smart India Hackathon
• Participant in the Google Generative AI Workshop
• Completed professional software and analytics simulations with British Airways and Deloitte

• Email: vedantagarwal039@gmail.com
• LinkedIn: linkedin.com/in/vedant-agarwal-36bb18142
• Portfolio: vedant-agarwal-812.vercel.app
• Kaggle: kaggle.com/vedantagarwal0812
• LeetCode: leetcode.com/u/Vedag812