Name : Ardhi Gagan
Alias : ardhigagan
Education : B.Tech Computer Science & Engineering — KIIT University (2023–2027)
CGPA : 9.30 / 10
Focus : AI Engineering · Full Stack Development · ML Research
Interests : LLM Systems · RAG Pipelines · Computer Vision · Data Engineering
Currently : Building AI-driven applications & seeking internship / entry-level rolesI am a final-year Computer Science student at KIIT University with a deep focus on building production-grade AI systems. My work spans the full spectrum — from fine-tuning transformer architectures and designing RAG pipelines to shipping full-stack web applications with real users. I treat engineering as a craft: clean abstractions, measurable outcomes, and systems that scale.
I bring a product engineering mindset to every project — not just making things work, but making them matter. Whether it's reducing legal document review time by 60%, winning national-level hackathons, or ranking in the top 0.4% in competitive programming contests, I operate at the intersection of rigor and impact.
Open To: AI/ML Internships · Software Engineering Roles · Data Science Positions · Full Stack Engineering · GenAI Engineering
Languages
Frontend
Backend & Databases
Cloud, DevOps & Tooling
AI / ML & Data Science
LangChain FAISS ChromaDB HuggingFace Groq LLaMA-3.3 sentence-transformers Legal-BERT LoRA LightGBM BART Pandas NumPy Matplotlib Streamlit
Visualization & Analytics
Tableau Power BI Matplotlib Seaborn
CS Fundamentals
Data Structures & Algorithms OOP DBMS Operating Systems Computer Networks System Design
| Domain | Proficiency | Details |
|---|---|---|
| Large Language Models | ██████████ Advanced | Prompt engineering, multi-agent pipelines, LLM orchestration via LangChain |
| RAG Systems | ██████████ Advanced | FAISS, ChromaDB vector stores; chunking strategies; semantic retrieval |
| Computer Vision | █████████░ Advanced | U-Net CNN segmentation, CLAHE preprocessing, boundary detection |
| NLP & Text Mining | █████████░ Advanced | BART summarization, Legal-BERT + LoRA fine-tuning, zero-shot classification |
| Gradient Boosting | ████████░░ Proficient | LightGBM for geospatial & temporal prediction tasks |
| Full Stack AI Apps | ██████████ Advanced | React + Flask + MongoDB; cloud-deployed AI products on GCP & Vercel |
| Data Engineering | ████████░░ Proficient | ETL pipelines, MongoDB Atlas, REST API design, JWT auth |
| MLOps & Deployment | ████████░░ Proficient | Google Cloud Run, Docker, Vercel, Render |
TRAVELMAiT — AI Travel Platform for Odisha
An intelligent full-stack travel platform built for Smart India Hackathon 2025, focused on Odisha's tourism landscape. Combines a semantic RAG pipeline with real-time third-party APIs to deliver personalised, context-aware travel recommendations.
| Attribute | Detail |
|---|---|
| Stack | React + Vite · Flask · MongoDB Atlas · ChromaDB · Groq / LLaMA-3.3-70B · sentence-transformers · Foursquare API · Cloudinary |
| Scale | Top 50 teams nationally — Smart India Hackathon 2025 |
| Architecture | Full RAG pipeline with semantic chunking and vector retrieval; JWT auth; mood-based trip planner |
| Integrations | Foursquare (hotels/restaurants), Cloudinary (photo CDN), Groq inference |
| Deployment | Vercel (frontend) · Render (backend) |
| Repository | github.com/ardhigagan/travelmaIt |
TRAVELMAiT is more than a travel guide — it is a domain-specific AI agent that understands Odisha's cultural and geographic nuances. The RAG pipeline retrieves semantically relevant destination data and passes it to an LLaMA-3.3-70B model for grounded, hallucination-resistant responses. Built with collaborator Anshuman Dev for SIH 2025, achieving national Top 50 recognition.
LegalLens v2.0 — NLP Contract Analysis Engine
An enterprise-grade legal AI system that automates contract review using transformer-based NLP. Reduces review time significantly while surfacing high-risk clauses with interpretable, structured output.
| Attribute | Detail |
|---|---|
| Stack | Python · LangChain · FAISS · Legal-BERT · LoRA · GPT-4o · BART · Streamlit |
| Performance | ~60% reduction in contract review time |
| ML Techniques | Long-doc chunking + BART summarization · Legal-BERT + LoRA fine-tuning · Zero-shot clause classification |
| Output | Risk heatmap dashboard · Clause-level risk scoring · Structured contract summaries |
| Security | On-premise vector store (FAISS); no raw document exposure |
| Repository | github.com/ardhigagan/legallens |
LegalLens v2.0 pipelines lengthy contracts through a chunking-and-retrieval system before running domain-adapted Legal-BERT for clause classification. LoRA fine-tuning allowed efficient adaptation with minimal compute. The result is a Streamlit dashboard that gives legal reviewers an instant, auditable risk overview — replacing hours of manual review with seconds of AI-assisted analysis.
Optic Disc Detection & Segmentation Pipeline
A medical imaging research pipeline that benchmarks deep learning against classical unsupervised methods for optic disc segmentation in retinal fundus images. Research paper co-authored and in progress.
| Attribute | Detail |
|---|---|
| Stack | Python · OpenCV · U-Net CNN · Scikit-learn · K-Means · DBSCAN |
| Performance | 92.17% Mean Dice Coefficient · 99.76% Pixel Accuracy (U-Net) |
| Techniques | Supervised CNN (U-Net) vs. unsupervised clustering (K-Means, DBSCAN) |
| Preprocessing | ROI-CLAHE · Mathematical elliptical fitting for zero-training boundary detection |
| Impact | Co-authored research paper (in progress); contributes to accessible retinal diagnostics |
| Repository | github.com/ardhigagan/optic-disc-seg |
This pipeline establishes a rigorous comparative study between a trained U-Net architecture and unsupervised clustering baselines. The unsupervised path uses targeted preprocessing — ROI-CLAHE contrast enhancement and elliptical fitting — to achieve competitive boundary localisation without any labelled training data, making it viable for low-resource clinical settings.
KaaryaAI — AI Chief of Staff (Multi-Agent Productivity App)
A multi-agent AI productivity system built for the Vibe2Ship Hackathon 2026 (Coding Ninjas × Google for Developers). Integrates Gmail, Google Calendar, and Google Drive into a single intelligent workflow orchestrator.
| Attribute | Detail |
|---|---|
| Stack | Python · Multi-Agent Pipeline · Gmail API · Google Calendar API · Google Drive API · Google Cloud Run |
| Architecture | Multi-agent orchestration with domain-specific sub-agents per integration |
| Deployment | Google Cloud Run (serverless, production-deployed) |
| Hackathon | Vibe2Ship 2026 — Coding Ninjas × Google for Developers |
| Capability | Email triage · Meeting scheduling · Document retrieval · Cross-service task delegation |
| Repository | github.com/ardhigagan/kaaryaai |
KaaryaAI demonstrates end-to-end agentic design — a central orchestrator decomposes user intent and delegates to specialised sub-agents handling Gmail, Calendar, and Drive independently. Deployed on Google Cloud Run for zero-downtime serverless execution. Submitted as a complete production-ready build within the hackathon window.
| Recognition | Details |
|---|---|
| Smart India Hackathon 2025 | Top 50 nationally — TRAVELMAiT AI Travel Platform |
| AINCAT 2026 | AIR 1861 · Top 7% nationally · 93.53 percentile |
| Codequezt #30 | Rank 43 · Top 0.4% nationally |
| LeetCode 100 Days Badge 2026 | 450+ problems solved across DSA domains |
| Vibe2Ship Hackathon 2026 | Submitted KaaryaAI — full multi-agent AI app on Google Cloud Run |
| Google × Kaggle AI Agents Course | Completed 5-Day AI Agents Intensive · Google learning badges earned |
| Gridlock Hackathon 2.0 | Traffic demand prediction · LightGBM · Leaderboard score ~86% |
learning:
- Advanced agentic AI architectures & multi-agent orchestration
- System design for distributed ML inference
- Deep reinforcement learning fundamentals
building:
- TRAVELMAiT v2 — enhanced RAG pipeline & mobile-first redesign
- Open-source LLM tooling for domain-specific RAG
- Data engineering pipelines for real-world datasets
exploring:
- LLM fine-tuning on constrained compute (QLoRA, PEFT)
- Knowledge graph integration with vector retrieval
- MLOps best practices for production AI systems
open_to:
- AI / ML Engineering Internships
- Software Engineering (Full Stack / Backend)
- Data Science & Analytics Roles
- GenAI / LLM Engineering Positions
- Research Collaborations in NLP or Computer Vision
"The best engineers are not those who know the most — they are those who build what matters most."