Full-Stack Developer · AI & LLM Engineer · Machine Learning Enthusiast
## 🖼️ Portfolio Preview
I'm a Computer Science & Engineering student at KIIT University, focused on building real-world software products across Full-Stack Development, AI/ML, Generative AI, and LLM applications.
I enjoy going beyond tutorials and building complete systems — from architecture and database design to APIs, authentication, AI integration, deployment, and production workflows. My strongest engineering interests:
Full-Stack Web Development · Software Engineering · AI/ML · Generative AI · LLM Applications · RAG & Hybrid-RAG · AI Agents · Backend Engineering · Cloud & DevOps · Data Structures & Algorithms
Independent, self-directed engineering experience — everything below was designed, built, and shipped by me end-to-end.
I've built and deployed 8+ production applications, working extensively with Next.js, React, TypeScript, Node.js, Python, FastAPI, PostgreSQL, MongoDB, Prisma, Redis, Docker, AWS, Vercel, and Render. That work spans REST APIs, authentication systems, role-based authorization, payment integrations, background jobs, dashboards, and cloud deployments.
On the AI side, I've built applications around LLMs, RAG, Hybrid-RAG, vector databases, embeddings, knowledge graphs, LangChain, LangGraph, and query routing — moving beyond chat wrappers into real retrieval and reasoning systems. I also work with CI/CD using GitHub Actions, Docker, AWS EC2, and AWS ECR to automate testing, containerization, and deployment.
And underpinning all of it: 500+ Data Structures & Algorithms problems solved across LeetCode and Codeforces.
AI/ML isn't a bolt-on for me — it's a core track of how I learn and build.
- Python · NumPy · Pandas · Scikit-Learn · TensorFlow
- Regression, classification, model evaluation, data preprocessing
- ML pipelines & MLOps — making training repeatable and deployable
- LLMs & Prompt Engineering
- RAG & Hybrid-RAG
- Embeddings & Vector Search
- Knowledge Graphs
- LangChain & LangGraph
- AI Agents & Tool Calling
To understand how LLMs actually work internally (not as a buzzword), I implemented a GPT-2-style language model from scratch, including:
- A custom tokenizer
- Multi-head attention
- Autoregressive generation
Building it by hand taught me what the abstractions hide — and made every LLM application I've built since more deliberate.
I'm particularly interested in building practical AI systems that answer questions reliably — not just impressively. My Hybrid-RAG work is the best example.
Vector similarity alone misses explicit relationships and multi-hop context. So I built a hybrid retrieval system that combines multiple signals:
- Dense vector retrieval with ChromaDB for semantic similarity
- Knowledge graphs built with NetworkX, using LLM-based triple extraction to pull structured (subject → relation → object) facts from documents
- Query routing — an LLM decides whether a query needs vector search, graph traversal, or both
- Hybrid retrieval — vector + graph + hybrid paths merged into coherent context
The result: answers grounded in both semantic meaning and explicit relationships, enabling multi-hop reasoning that pure vector search can't handle.
I enjoy building complete products — not only frontend interfaces. From database schema to deployment, I care about the whole pipeline.
REST APIs · JWT authentication · Authorization · Role-Based Access Control
I'm interested in understanding how applications move from development to reliable production systems:
- Docker — containerizing applications for consistent environments
- GitHub Actions — CI/CD workflows for automated testing & deployment
- AWS EC2 / AWS ECR — container registry and cloud compute deployments
- Vercel · Render — managed deployment for web apps and services
- Linux — day-to-day comfort with the command line and servers
| Category | Technologies |
|---|---|
| Languages | TypeScript · JavaScript · Python · C++ · SQL |
| Frontend | React · Next.js · Tailwind CSS · shadcn/ui · Redux Toolkit · Framer Motion |
| Backend | Node.js · Express.js · Bun · FastAPI · REST APIs · JWT · RBAC |
| Databases | PostgreSQL · MongoDB · Prisma · Redis · Supabase · Neon · FAISS · ChromaDB |
| AI / ML | Scikit-Learn · TensorFlow · NumPy · Pandas · Model Evaluation · MLOps |
| LLM / GenAI | LangChain · LangGraph · RAG · Hybrid-RAG · Knowledge Graphs · AI Agents · Prompt Engineering |
| Cloud & DevOps | Docker · GitHub Actions · AWS EC2 / ECR · Vercel · Render · CI/CD |
| Developer Tools | Git · VS Code · Linux · Postman · Vite |
🛒 GoCart — Multi-Vendor E-Commerce Platform
A modern multi-vendor marketplace with distinct buyer, seller, and admin workflows. Tech Stack: Next.js · React · TypeScript · Redux Toolkit · Prisma · PostgreSQL · Stripe · Clerk · ImageKit · OpenAI SDK
- Built role-aware workflows for buyers, sellers, and admins on a shared, scalable Prisma + PostgreSQL data layer
- Integrated secure Stripe payments, Clerk authentication, and media management via ImageKit
- Added an AI-powered listing assistant (OpenAI SDK) to streamline product creation
- 🔗 Live
A multi-portal grocery delivery system covering the complete order lifecycle. Tech Stack: Node.js · Express · JWT · Stripe · Redis · Cloudinary · Inngest
- Built customer, store-admin, and delivery-partner portals with JWT authentication and role-based access control
- Integrated Stripe with webhook verification for trustworthy payments
- Implemented live order tracking, OTP delivery verification, Cloudinary media, and Inngest background jobs
🧠 HYBRID-RAG — Advanced AI Retrieval System
Combines vector search, knowledge graphs, and LLM-powered query routing. Tech Stack: Python · Gemini · ChromaDB · NetworkX · Knowledge Graphs · RAG
- Built dense vector retrieval with ChromaDB alongside knowledge-graph traversal via NetworkX
- Used LLM-based triple extraction to structure documents into graph facts, enabling multi-hop reasoning
- Implemented query routing that selects vector, graph, or hybrid retrieval per query
📄 RAG Chatbot — Document Intelligence
End-to-end document intelligence with Retrieval-Augmented Generation. Tech Stack: Python · FastAPI · LangChain · Groq · FAISS · Sentence Transformers · Docker
- Built a full document ingestion → chunking → embedding → retrieval → generation pipeline
- Grounded conversations in user-uploaded documents with semantic retrieval (FAISS)
- Dockerized and deployed to Render
- 🔗 Live
🤖 J.A.R.V.I.S. OS v4.0 — Autonomous AI Assistant
A voice-enabled assistant combining LLMs, RAG, web search, speech synthesis, and desktop automation. Tech Stack: Python · LangChain · Groq · FAISS · Sentence Transformers · Streamlit · Edge-TTS
- Orchestrated retrieval, web search, voice I/O, and desktop tools behind a unified interaction flow
- Maintained useful context while coordinating multiple AI and system tools
Machine learning applied to invoice analytics. Tech Stack: Python · Streamlit · Scikit-Learn · Plotly · Pandas
- Built freight-cost prediction and risk classification models on invoice data
- Delivered insights through an interactive Streamlit dashboard with Plotly visualizations
🧾 AI Resume Builder — Full-Stack AI Platform
An AI-powered platform for creating, enhancing, managing, and publishing professional resumes. Tech Stack: React · Node.js · Express · MongoDB · Redux Toolkit · Gemini API · ImageKit
- Synchronized editable structured content with AI-assisted writing suggestions
- Handled media management and publishable output end-to-end
- 🔗 Live
🔄 ML Pipeline with CI/CD — MLOps
An end-to-end ML pipeline with automated training, Dockerization, CI/CD, and AWS deployment. Tech Stack: Python · Scikit-Learn · Docker · GitHub Actions · AWS ECR · AWS EC2 · Flask
- Automated the path from data preparation and training to containerized cloud deployment
- Built a GitHub Actions CI/CD workflow pushing images to AWS ECR and deploying to AWS EC2
My engineering philosophy, in practice:
- Understand fundamentals first. I implemented a GPT-2-style model from scratch because I want to understand systems, not blindly use abstractions.
- Build things from scratch. Databases, auth, pipelines, retrieval — I learn by implementing before leaning on tools.
- Turn ideas into working products. Tutorials end where projects begin; I ship complete systems with real users in mind.
- Explore new tech through real projects. Every new technology I pick up gets applied to something that actually needs it.
- Continuously improve. Architecture and code quality are iterative — I refactor as I learn better designs.
500+ problems solved across LeetCode and Codeforces, covering:
- Data Structures & Algorithms
- Graphs
- Dynamic Programming
- Greedy Algorithms
- Problem Solving
Alongside CS foundations: Operating Systems · DBMS · Computer Networks · OOP.
- Deepening my understanding of advanced AI/LLM engineering and RAG systems
- Exploring AI agents and agentic workflows
- Building toward production-grade full-stack architecture and system design
- Improving backend engineering and MLOps practices
- Learning cloud infrastructure for reliable deployments
I'm interested in opportunities involving:
Software Engineering · Full-Stack Engineering · Backend Engineering · AI Engineering · Machine Learning Engineering · LLM / Generative AI · AI Application Development
I'm interested in collaborating on AI/ML projects, LLM applications, developer tools, full-stack products, open-source projects, and interesting engineering experiments — especially where the hard part is the system design, not the tech stack.
Let's Connect 🤝
🌐 Portfolio · 💻 GitHub · 💼 LinkedIn · 📸 Instagram · 📧 Email
🌐 The Portfolio is the best place to start — it features my latest projects and professional information.
A high-end, professional developer portfolio built with TanStack Start + shadcn/ui. Live site: https://suraj-builds-ai.lovable.app
- Dark-first premium design — Vercel/Linear-inspired minimal aesthetic with charcoal surfaces, subtle borders, and blue/violet accents; polished light mode included
- Sticky responsive navigation — section links, social icons, resume download, theme toggle, and an accessible mobile menu
- Project case studies — each project card opens a detailed dialog with overview, problem, solution, key features, engineering considerations, and tech stack
- Animated stat counters — subtle number animations that respect
prefers-reduced-motion - Skills taxonomy — category tabs across Frontend, Backend, Databases, AI & LLM, Machine Learning, Cloud & DevOps, Programming, and CS Fundamentals
- Journey timeline — visual timeline from 2021 to present with the current milestone highlighted
- Validated contact form — opens a mail draft; no fake sending
- SEO & accessibility — semantic HTML, Open Graph metadata, keyboard navigation, focus states, and reduced-motion support
# Install dependencies
npm install
# Start the dev server
npm run dev
# Build for production
npm run build├── public/
│ └── resume.pdf # Downloadable resume
├── src/
│ ├── components/
│ │ ├── portfolio/ # Page sections & reusable components
│ │ └── ui/ # shadcn-style UI primitives
│ ├── data/
│ │ └── portfolio.ts # Single source of truth for all content
│ ├── hooks/ # Motion & utility hooks
│ ├── lib/ # Shared utilities
│ └── routes/ # File-based routes
└── vite.config.ts
All portfolio content (profile, projects, skills, milestones, stats) lives in src/data/portfolio.ts — edit that one file to update the site.
© 2026 Suraj Kumar Gupta · Built with TanStack Start + shadcn/ui