I live in the space between research and real software — comfortable training a model and shipping the service around it.
Available🏆 IEEE Best Paper🎓 Columbia👋
I'm an M.S. Computer Science student at Columbia University who likes taking AI from prototype to production. I've shipped core banking software at Oracle Financial Services, built a multi-agent compliance validation system on Azure at atQor, and published 6 research papers (including an IEEE Best Paper Award). I move comfortably between deep LLM/agentic work and real software engineering — APIs, pipelines, and the unglamorous delivery lifecycle that makes things actually ship.
Education
Columbia University
Expected Dec 2027
M.S. in Computer Science
Focus on production-grade AI and LLM systems.
Symbiosis Institute of Technology
Sep 2021 – Jun 2025
B.Tech, Computer Science (Honors in AI & ML)
3rd Rank in Department (Top 2%) · GPA 9.10/10 · Merit Scholar.
Symbiosis Institute of Business Management
Mar 2022 – Jul 2023
Diploma in Business Management
Batch Topper · GPA 9.17/10.
Anandalaya School
2018 – 2021
AISSCE (CBSE), Class XII & X
Class XII: 95.6% · Class X: 96% (Merit Certificate in Mathematics).
git log --experience
Where I've shipped
Banking software at Oracle, a multi-agent compliance system at atQor, and research along the way.
ORCL
Associate Consultant
Oracle Financial Services Software
Jun 2025 – Jun 2026Pune, India · Full-time
Built and customized 22 RESTful APIs, 9 screens, and 2 GEFU interfaces in Oracle FLEXCUBE Retail Banking (FCR) across CASA, Loans, and Deposits using Java, PL/SQL, and Oracle JET.
Owned end-to-end delivery of 33 components: development, functional validation, post-validation fixes, UAT support, and Pre-Production release creation & deployment.
Communicated directly with client stakeholders to resolve UAT conflicts; packaged and released UAT builds near-daily and Pre-Production builds 1–2x weekly via Jenkins for 3+ months.
JavaPL/SQLOracle JETJenkinsREST APIsFLEXCUBE
atQ
Associate Data Scientist Intern
atQor
Jan 2025 – May 2025Ahmedabad, India · Internship
Built a task-specific multi-agent compliance validation framework (Semantic Kernel) with rule-based and LLM-driven agents for matching, name, and wage checks — cutting validation time from hours to minutes.
Developed a PDF-to-JSON pipeline into Azure Cosmos DB with dynamic rule configs and async FastAPI endpoints (Pydantic-structured outputs), delivering sub-2s latency and ~25% higher reliability.
Deployed on Azure for UAT; ran end-to-end and regression testing and refined logic from client feedback.
Prototyped 3 GenAI tools: ColPali + GPT-4o PDF Q&A, a real-time speech assistant, and a RAG legal copilot.
Benchmarked YOLOv5 vs. YOLOv7 for tomato leaf disease detection and classification for an agricultural-robot use case.
Computer VisionYOLOPyTorch
GK
Service Learning Project Lead
GramKalyan NGO & Varhad Capital
Jul 2023 – Oct 2023Pune, India · Leadership
Led a 5-member team to build and deliver 2 websites for the NGO and its parent organization, improving digital outreach.
WebLeadershipReact
ls ./projects
Things I've built
Shipped products, award-winning hacks, and published systems — spanning LLMs, ML, and computer vision.
featured
Multimodal RAG for Multilingual Product Support
Jul 2024 – Nov 2024
A RAG chatbot for product support in English, Hindi, and regional Indic languages. Benchmarked embeddings (IndicBERT, mT5) stored in ChromaDB and multiple retrieval strategies over technical PDFs, with language-aware text, voice, and visual responses. A ColPali + Vision-Language-Model approach won on interpreting visual document content.
Published · IAJIT
PythonRAGIndicBERTmT5ChromaDBColPali
featured
Sight Beyond Sight
Feb 2023 – Mar 2023
A web-accessibility tool for the visually impaired: parses a page with BeautifulSoup, summarizes it via OpenAI, narrates it through Azure TTS in a natural voice, describes images with Google Vision, and answers questions through a chatbot — with Redis caching results for 3 hours. Top-voted submission at OpenAI Stack Hack among 7,000+ participants.
An AI emotion-analysis dashboard for psychological well-being that analyzes text and images from social media. Hybrid transfer learning fuses BERT (text) and Vision Transformers (images); trained on GoEmotions and AffectNet, it reaches 89.7% classification accuracy, outperforming unimodal baselines. Published in IJAI.
89.7% accuracy · Published
PythonBERTVision TransformersTensorFlowMultimodal
AI-Powered Malware Detection & Classification
Sep 2024 – Nov 2024
An ML malware-detection system classifying Portable Executable malware with Random Forest (99.4%) and polymorphic URLs with Logistic Regression (97.85%), using TF-IDF feature transformation. Handles class imbalance via SMOTE and random oversampling inside a modular pipeline that automates feature extraction, classification, and real-time detection.
99.4% / 97.85% accuracy
PythonScikit-learnRandom ForestTF-IDFSMOTE
Networking Blog — DevSecOps Deployment
Sep 2024 – Nov 2024
A secure, automated, and scalable web app deployed through a full DevSecOps pipeline: every commit triggers a GitHub → Jenkins CI/CD flow with SonarQube, OWASP, and Trivy security scanning. Containerized with Docker Compose and provisioned on AWS EC2 via Terraform for reproducible, scalable infrastructure.
CI/CD · security-scanned
JenkinsAWSDockerTerraformSonarQubeTrivy
CropPal — Crop & Wheat Growth Detection
Jan 2024 – May 2024
A computer-vision system that identifies crops and detects wheat growth stages using transfer-learned CNNs, built to optimize agricultural insurance workflows, precision agriculture, and supply-chain monitoring through accurate crop assessment.
95%+ accuracy
PythonCNNsTransfer LearningFastAIReact
AURA — Emotion-Based Music Recommendation
Apr 2024
An AI music-recommendation system that reads a user's emotion in real time from facial expressions using CNNs, then curates personalized playlists via the Spotify API — bridging music discovery and emotional intelligence in a real-time, user-friendly interface.
real-time emotion → music
PythonCNNOpenCVSpotify API
MedSync — Unified Patient Treatment Platform
Oct 2023 – Nov 2023
A healthcare management platform with a Unified Patient Treatment Platform (UPTP) to securely store and share medical records, diagnoses, and prescriptions, plus multi-source health-data integration and online appointment booking. Awarded 3rd place in the Healthcare track at the IEEE India Council Hackathon 2.0 (2023).
🏆 3rd · IEEE IC Hack 2.0
HTMLCSSJavaScriptHealthcare
OmniVision — Image Captioning
Aug 2022 – Oct 2022
An image-captioning model built with Keras and TensorFlow that generates one-line descriptions of images, created to assist the visually impaired. Developed during the _VOIS Innovation Marathon 2022, where the team ranked 11th out of 540.
Ranked 11 / 540 teams
PythonKerasTensorFlowCNN
./mudit --interactive
Talk to my terminal
A real shell, not a GIF. Type commands, use ↑/↓ for history, Tab to autocomplete. Go on, poke around.
bash — mudit@portfolio
mudit@portfolio:~$
cat ./research.bib
Published research
Six peer-reviewed publications across RAG, multimodal AI, ML security, and cryptography — including an IEEE Best Paper Award.
2026
Optimizing Multimodal RAG Systems for Multilingual Product Support
Intl. Arab Journal of Information Technology (IAJIT), Vol. 23, No. 3