Projects
Products and builds by Atharva Sonar.
These are the products and systems I've designed and built: Velquix products, hackathon builds with team Deploy House, and earlier projects. Each case study explains the problem, how the system works, its architecture and its current status.
Featured projects
SpendDecode
AI Fintech
A personal finance platform that helps people understand their transactions, analyse spending habits, find saving opportunities, and make better purchasing decisions — with an AI layer that explains the numbers instead of just charting them.
Velquix Learn
EdTech · Learning Platform
A developer learning platform that turns scattered tutorials into a clear path: career roadmaps, curated free resources and playlists, practice projects, and progress tracking.
ClaimBlitz
Multi-Agent AI · OFFGRID 1.0
A medical insurance claim processor where specialised agents extract, validate, analyse and decide. When analysts disagree, a structured debate and a judge agent settle the verdict, and low-confidence cases are escalated to a human.
SafeHer AI
AI · Safety Technology · Build And Grow: AI Hackathon 2.0
A full-stack safety platform that tracks live route context, scores risk with a scikit-learn engine, listens for spoken SOS phrases, and lets an agentic engine recommend or trigger protective actions as risk rises.
FitHub Gym
Full-Stack · AI
A Django gym platform covering memberships and subscriptions, trainer dashboards and workout videos, plus an AI trainer that counts biceps curls from the webcam using OpenCV and MediaPipe pose tracking over WebSockets.
Earlier work
- Finexa (GHRhack 2.0) — An earlier AI personal-finance platform: document-based expense extraction, a financial health score, an AI coach, goals and habit challenges. Much of what I learned here shaped SpendDecode. Source on GitHub
- FinRight AI (MumbaiHacks 2025) — A full-stack financial intelligence platform focused on personal liquidity and group expense sharing. Source on GitHub
- RAG PDF Assistant — A Streamlit RAG app: PDF extraction, overlapping chunks, local sentence-transformer embeddings, FAISS search and Gemini answers grounded in the retrieved text. Source on GitHub