I'm a Generative AI Developer at Karanji Infotech Pvt Ltd, where I design and ship LLM-powered automation systems that are actually in production, not just prototypes. My focus is Retrieval-Augmented Generation (RAG), prompt engineering, and full-stack AI products that people can rely on.

My flagship project, ComplianceIQ, is a citation-enforced RAG assistant for RBI/SEBI/IRDAI regulatory documents every answer is grounded in a retrieved source and evaluated with RAGAS. I care about the difference between an LLM that sounds right and one that's verifiably right, and that distinction shapes how I design retrieval pipelines, prompts, and evaluation loops.

Before GenAI, I trained as a VR/Unity developer where I picked up an instinct for real-time, interactive systems that I now bring to AI product design.

• 🎬 AI Storyboard Generation Platform : Gemini 3.1 Pro + LangChain pipeline automating scene descriptions, character dialogue, and visual prompts, cutting storyboard creation effort by ~70%.
• 📚 E-Learning Script & Module Generator (ESLM) : Multi-step prompt chaining with Gemini 3.1 Pro and structured output, letting non-technical users generate course-ready content in under 15 minutes (~80% reduction in instructional design effort).
• 📊 Progress Tracker Platform : Full-stack project tracking tool (Next.js, Supabase, Node.js, Google Cloud Run) for monitoring project progress and team coordination.
• ⚙️ CI/CD Pipelines : GitHub Actions workflows automating build, test, and deploy for cloud-hosted applications.

These are internal/proprietary systems built for my employer, so there's no public repo happy to walk through the architecture in an interview.

Docker · Agentic AI workflows · Production deployment & observability for LLM apps

Generative AI / LLM

Backend

Frontend

Databases

DevOps / Cloud

Languages

A RAG assistant that refuses to answer questions it can't ground in a retrieved source built for RBI/SEBI/IRDAI regulatory documents.

Problem it solves: Regulatory and compliance teams can't afford an AI assistant that hallucinates a rule that doesn't exist. ComplianceIQ enforces page-level citations on every answer and blocks responses that fall outside the ingested document corpus.

Tech stack: Python · FastAPI · Next.js/TypeScript

AI/LLM technologies: Hybrid retrieval (BM25 + vector embeddings) · ChromaDB · Groq-hosted Llama 3.3 · forced-grounding prompt design · RAGAS evaluation (faithfulness, answer relevancy, context precision) against a hand-curated golden question set

Key impact: Every answer is traceable back to a specific page in a regulatory circular no ungrounded output ships.

🔗 Repo: github.com/pavith-raj/ComplianceIQ 🌐 Live demo: coming soon

A mobile-first salon management app with computer-vision-powered beauty recommendations.

Problem it solves: Salons need both efficient booking/queue operations and a way to offer personalized recommendations most tools do one or the other.

Tech stack: React Native · Node.js · MongoDB · Python

AI/LLM technologies: Computer vision (OpenCV) for skin analysis and hairstyle prediction, feeding a personalized cosmetic product recommendation engine

Key impact: Full-stack mobile app covering appointment booking, real-time queue management, JWT authentication, and RBAC end to end, not just the AI layer.

🔗 Repo: github.com/pavith-raj/AuraX 🌐 Live demo: not yet public

A web app that automates dynamic question paper creation from a managed question bank.

Problem it solves: Manually assembling exam papers is slow and inconsistent; this system generates papers dynamically from a structured question bank.

Tech stack: React.js · Node.js · MongoDB

Key impact: RESTful APIs and efficient question bank management, built following an Agile-based SDLC.

🔗 Repo: github.com/pavith-raj/question-paper-generator

• Retrieval Augmented Generation (RAG) & Agentic AI — DeepLearning.AI
• Software Engineering & Agile Software Development — Infosys Springboard
• Python Foundation Certification — Infosys Springboard
• Object-Oriented Programming using Python — Infosys Springboard
• Database Management Systems — Infosys Springboard
• Introduction to Model Context Protocol (MCP)