CSE @ RNSIT · Bengaluru, India
I build systems for problems where a basic API + chatbot isn't enough.
|
|
|
My approach: find an under-explored problem → build the hard part → measure it → turn it into a usable system.
|
Municipal Decision Intelligence Multimodal citizen reports → classification → duplicate clustering → severity → explainable reasoning → department routing → GIS operations. Stack: Gemma · JavaScript · Multimodal AI · GIS |
TinyML Environmental Intelligence Environmental audio classification deployed directly to ESP32. 84% INT8 accuracy · 69.1 KB model · 17 classes Stack: Python · CNN · TFLite · ESP32 · MFCC |
|
AI-Assisted Personal Safety Trip monitoring, threat detection, saferoutes, emergency escalation, SOS, voice interaction and geospatial safety intelligence. Stack: React · Flask · Scikit-learn · Leaflet · Docker |
Neural Lost & Found Privacy-first semantic matching for physical-world lost & found reports. Stack: TypeScript · Generative AI · Semantic Search |
|
Personal AI Architecture Retrieval, context, memory and reasoning components for a modular personal AI system. Stack: Python · RAG · LLMs |
Video-Grounded Learning Retrieves relevant course-video context and generates answers grounded in source material and timestamps. Stack: Python · BGE-M3 · Qwen3 · RAG |
|
69.1 KB INT8 TinyML model |
84% EcoSentinel INT8 accuracy |
17 environmental classes |
60M+ rows handled in FX research |
- 🏆 EcoSentinel — 1st Prize, NextGen Project Expo — RNSIT
- 🥈 EcoSentinel — Runner-up / 2nd Prize, INoVEX 3.0 — IIT Dharwad
- 🏅 EcoSentinel — Semifinalist, Build With Bharat 2.0 — NIT Delhi
- 🏆 FeelSafe — 1st Prize, Rajarajeshwari College of Engineering — Bengaluru
- 🏅 Arbitrage Arena 2026 — Semifinalist, IISc Bengaluru
Large-scale ML
├── streaming / chunked data
├── memory-efficient feature engineering
└── scalable model training
Multimodal Intelligence
├── video reasoning
├── evidence extraction
└── uncertainty-aware retrieval
Edge AI
├── TinyML
├── quantized inference
└── sensor fusion
Agentic Systems
├── tool-using agents
├── RAG
└── long-term context