I'm Jaswanth Sanjay (Nekkanti Jaswanth Sanjay) — a B.Tech Computer Engineering student at Parul University, Vadodara, building on-device AI systems, privacy-first apps, and agentic tools from the ground up. I care about software that runs locally, keeps data private, and doesn't depend on the cloud to be useful.
⚡ I like taking ideas from "can this even run on a phone?" to a working app — model inference, UI, and architecture, end to end.
🎓 Certified in Oracle Cloud GenAI, AWS Data Engineering, and Hugging Face Agents.
BIT — Offline Android LLM chat app. Founder/sole developer. Material 3 design with a custom ModalBottomSheet toolbar, SharedTransitionLayout animations, liquid-glass surfaces (Haze), and tool-call trace UI with animated status pills — all running local inference, no cloud round trip.
llama.kt — A Kotlin-native on-device LLM inference engine, built for the Tata Technologies InnoVent-27 hackathon ("AI at the Edge"), targeting Industrial, Automotive, and Aerospace edge deployment scenarios.
ShadowText — Zero-knowledge, end-to-end encrypted messaging app. Tor for transport privacy, TweetNaCl for crypto, Next.js frontend, Supabase/Firebase backend. Built around a "the server should never be able to read your messages" philosophy.
AskAtharva — RAG-based assistant using Ollama/LLaMA-3 with a Firestore vector store and a custom card-based chat UI.
Atharva MCP Agent — A local LLaMA-3 agent wired up over MCP with email, calendar, and Perplexity-style web search tools — one of my early deep dives into agentic tool-use architecture.
AI Call Assistant — Android app (MVVM, Room DB, Google Cloud APIs) for call handling and assistance.
SAR-Optical Fusion Engine — Cloud-removal pipeline for satellite imagery combining Sentinel-1/2 data with diffusion models/GANs (U-Net, PyTorch, GDAL), built for ISRO BAH 2026.
Core interests: on-device LLM inference, MCP-based agentic workflows, privacy-first architecture, cross-platform delivery (Android, web, Telegram).
BIT (offline Android AI chat app) · llama.kt (Kotlin-native edge inference engine) · ShadowText (zero-knowledge E2E messaging) · fine-tuning dataset for Atharva AI
"Building the ASI is Future Goal."