I build production AI systems. LLM pipelines. RAG architectures. Scalable backends.
Not prototypes. Not notebooks. Systems that ship.
I architect intelligent systems — from autonomous AI agents with multi-service orchestration, to GAN-based image enhancement models, to legal domain LLMs with prompt injection defense. I also build Android applications currently live on the Play Store.
Previously: AI Intern @ Broadrange AI (Chicago) · Android Developer @ Techpuram (Madurai)
Education: B.Tech (Hons) CSE @ RV University · AI/ML Major · FinTech Minor · Class of 2027
Python Java C++ Kotlin C SQL
TensorFlow PyTorch Keras scikit-learn LangChain FAISS
Spring Boot Flask Jetpack Compose REST APIs Microservices
PostgreSQL MySQL SQLite pgvector Docker Git Linux Ollama
◼ LLM SERVICE MCP — Autonomous AI Agent System
Microservice-based AI system with dynamic tool orchestration across RAG, Web Search, and Chat services. BERT-based intent classification handles real-time prompt routing and multi-step reasoning.
BERT RAG LLM Spring Boot Microservices Intent Classification
◼ GRAPH RAG — Knowledge Graph + Vector Search + LLM
Hybrid retrieval system combining NetworkX knowledge graphs with FAISS vector similarity and Groq LLM. Interactive D3.js graph visualization. Document upload with intelligent chunking.
NetworkX FAISS Groq D3.js Flask
◼ FACE SUPER RESOLUTION — Real-ESRGAN Enhancement
Fine-tuned Real-ESRGAN using transfer learning and custom loss balancing on paired HR-LR datasets.
| PSNR | SSIM |
|---|---|
| 29.15 → 33.95 dB | 0.8985 |
Real-ESRGAN PyTorch Transfer Learning GANs
◼ NL-TO-SQL — Offline AI Database Manager
Converts natural language into executable SQL using locally hosted Mistral via Ollama. Schema-aware prompting. SQL validation. Snapshot-based rollback. Human-in-the-loop execution.
Ollama Mistral SQLite NL-to-SQL
◼ LEGAL AI LLM — Built at Broadrange AI, Chicago
Legal domain LLM trained on curated datasets to minimize hallucination. Document ingestion with chunking. PromptGuard V2 for prompt injection defense. RAG architecture for contextual grounding.
LLM RAG PromptGuard V2 Document Ingestion
◼ BRAIN TUMOR SEGMENTATION — Medical Imaging
U-Net encoder-decoder architecture for semantic segmentation of brain tumors from MRI scans. Built for Kaggle competition.
U-Net TensorFlow Medical Imaging
◼ GEO GPS CAMERA — 100+ Play Store Downloads
Production Android app. Secure location-tagged image capture. Server-side validation against GPS spoofing. Real-time address overlay via Google Maps API.
Kotlin Jetpack Compose CameraX Google Maps API
◼ PERSONAL ASSISTANT — Fully Offline Voice AI
Siri-like voice assistant for macOS. No cloud dependency. Local speech recognition, local LLM, multi-turn conversations.
Speech Recognition Local LLM macOS Python
MORE PROJECTS
| Project | What It Does |
|---|---|
| INTELLI-CHAT-AI | Smart Document & Web Assistant — RAG + LLM |
| TRAVEL-PLANNER-MCP | AI Travel Planning System — MCP Architecture |
| Currency-Exchange | Currency Exchange Application |
| Fake-News-Check | Fake News Detection — ML + NLP |
| Movie_Reccomendation_System | Moodix — TF-IDF + Cosine Similarity |
| Bank_Statement_Organizer | Bank Statement Analysis — FinTech |
| Blackchain-Voting-System | Blockchain Voting — Solidity |
| WEATHER_SONG_RECOMMENDATION | Weather-Based Song Recommender |
| Print-Quality-Dataset-Generator | Print Quality Dataset — Computer Vision |
Artificial Intelligence Intern — Broadrange AI · Chicago, Illinois
Jun 2025 — Aug 2025
Built Legal Domain LLM. RAG architecture. Document ingestion with chunking. PromptGuard V2 for prompt injection. Minimized hallucination through curated legal datasets.
Android Native App Developer — Techpuram Technology · Madurai, Tamil Nadu
Feb 2025 — Aug 2025
Built Geo GPS Camera App. 100+ Play Store downloads. Jetpack Compose + Kotlin + Google Maps API. Server-side validation against GPS spoofing.
B.Tech (Hons) Computer Science — RV University, Bangalore 2023–2027
Major: AI & Machine Learning · Minor: FinTech
Best Project Award — Structured Innovation
NPTEL — Programming in Modern C++ (2024–2025)
I don't train models. I ship systems.
31 repositories · AI/ML · Backend · Android · FinTech
