Madhav zanwar

B.Tech CSE (AI & ML) student | Full-stack + AI builder | Open source contributor

Open Source Contributor · GirlScript Summer of Code (GSSoC)Pimpri Chinchwad College of EngineeringPune, Indiaopen to internships always.

Open to roles

CommunicationGITProblem SolvingProject Management

GitHub LinkedIn

Brand & Creative Builder Award '26

Project Background & Objective: We are living through an epidemic of sensory overload. Urban noise pollution and relentless digital notifications are leading to record levels of burnout. While the wellness industry focuses on internal solutions (meditation apps), NEMOS addresses the physical environment. The objective is to create a premium, spatial wellness brand that transforms overstimulating environments into biophilic, sensory-inclusive sanctuaries. Intended Audience: Primary: High-performing urban professionals suffering from digital fatigue. Secondary: Neurodivergent individuals requiring sensory-friendly, highly controlled physical environments to decompress. Central Brand Idea & Creative Strategy: "The Architecture of Quiet." The core strategy relies on visual and sensory contrast. We contrast the sharp, jagged noise of the modern world with the soft, organic, adaptive quiet of a Nemos space. By visualizing the invisible (noise and stress) and building an interactive system to neutralize it, the brand acts as a spatial healer. Visual Identity & Practical Applications: The visual identity transitions from harsh "Noise" palettes (Concrete, Charcoal) to grounded "Silence" palettes (Sage Green, Terracotta). Interactive Demo (Built in React): A live web prototype showcasing the digital-to-physical integration of the NEMOS ecosystem and state control. OOH Campaign ("Quietest Billboards"): High-contrast subway ads constructed from sound-absorbing mycelium, creating literal quiet zones. Product Packaging: Zero-waste, molded mycelium unboxing experiences for the "Sanctuary Kit." Process & AI Disclosure: This project bridges the gap between theoretical branding and technical execution. The brand positioning, UX logic, and strategy are entirely original. Midjourney v6 was utilized for initial architectural environments and material textures, which were then composited and color-graded. The brand identity, typography systems, and UI components are original vector assets. Google AI Studio was utilized to assist in structuring the React components, Tailwind state transitions, and interactive UI logic.

2 media files · nemos-architecture-of-quiet.vercel.appView

PCA Face Recognition using MATLAB

Developed a face recognition system in MATLAB using Principal Component Analysis (PCA) and the Eigenfaces algorithm for dimensionality reduction and facial feature extraction. The project processes grayscale facial images, computes the mean face, generates eigenfaces, projects images into eigenspace, and recognizes faces by comparing Euclidean distances in the reduced feature space. Key Features: • Mean face computation • Eigenface generation and visualization • Face reconstruction from principal components • PCA-based face recognition • Euclidean distance matching The implementation achieved a recognition accuracy of 93.75% on the ORL Face Dataset and demonstrates the application of classical machine learning techniques in computer vision.

github.comView

AI Placement Intelligence Platform

Built a full-stack AI-powered placement preparation platform that helps engineering students assess their placement readiness through resume analysis, skill-gap detection, and AI-based mock interviews. The platform analyzes uploaded resumes, extracts technical skills, identifies missing technologies, calculates a placement readiness score, and generates personalized preparation suggestions. It also includes a mock interview module and a dashboard that visualizes student strengths and improvement areas. Designed with scalability in mind, the architecture is ready for integration with modern Large Language Models (LLMs) to provide intelligent resume reviews and adaptive interview feedback. Key Features: • Resume analysis and skill extraction • Placement readiness scoring • Skill gap detection • AI mock interview simulation • Personalized preparation roadmap • Student analytics dashboard Tech Stack: Frontend: HTML, CSS, JavaScript Backend: Python, Flask Deployment: Render This project demonstrates full-stack development, backend API design, AI product architecture, and real-world problem solving for placement preparation.

2 media files · github.comView

MemoryVerse AI '26

1. Live app: upload a PDF → auto-classified (category, skills, summary) via one Gemini call. 2. Search tab: natural-language queries ("show my AI projects") ranked by semantic similarity, not keywords. 3. Browse tab: all docs auto-sorted by category. Timeline tab: chronological view from extracted dates. 4. Relationships: `/api/relationships/build` — one batched Gemini call finds cross-document connections (not per-pair, avoids O(n²) calls). 5. **Cold start note**: backend is on Render free tier — first request after inactivity can take 30–60s. Not a bug. 6. **Design choice**: single Gemini call per doc (classify + extract + tag skills together) — keeps latency low, stays in free-tier limits. 7. **Design choice**: pure-Python cosine-similarity vector search instead of ChromaDB — its compiled dependency broke on Windows mid-build; swapped deliberately for reliability under deadline. Fine at student-scale (tens of docs), documented as a tradeoff. 8. Originally Railway-hosted; migrated to Render + Vercel after Railway's trial expired mid-build. 9. Local setup, full API docs, and architecture diagram are in the repo README. 10. Limitations stated upfront: no auth (single-user demo scope), relationship engine is one LLM pass not a formal knowledge graph, dates depend on being present in source text. 11. Solo build, ~48 hours: backend, AI integration, frontend, deployment all done end-to-end.

4 media files · memoryverse-ai-wheat.vercel.appView

QuickSketch — AI Sketch Recognition

QuickSketch is a full-stack AI web app that recognizes hand-drawn sketches in real time. Users draw on a canvas, and a custom-trained convolutional neural network predicts what they've drawn from 16 categories — apple, bicycle, camera, sun, pizza, umbrella, star, book, fish, crown, diamond, lightning, guitar, house, mountain, and chair. The model is a PyTorch CNN trained from scratch, achieving 95.6% validation accuracy. The backend is built with Flask and deployed on Railway; the frontend is a static canvas-drawing interface deployed on GitHub Pages. Built as the flagship project from a focused two-week self-directed learning sprint covering deep learning fundamentals and full-stack deployment — from model architecture and training through to a live, publicly usable product.

3 media files · madhavzanwar.github.ioView

MemoryVerse AI — Intelligent Digital Identity System

Every student accumulates certificates, resumes, project reports, and internship letters across years — scattered across folders, emails, and drives. MemoryVerse AI transforms this fragmented data into a structured, searchable, intelligent digital identity. Built for MemoryVerse AI '26, the system uses a single Gemini API call per document to classify it (Project / Skill / Certification / Internship / Achievement / Academic) and extract structured metadata — title, date, skills, and a summary. Documents are embedded using Gemini's embedding model and stored in a lightweight cosine-similarity vector store, enabling true semantic search: a query like "show my AI projects" matches on meaning, not keywords. A batched relationship-inference pass connects documents across the corpus (e.g. a certification's skills resurfacing in a later project), and a chronological timeline view is generated automatically from extracted dates. Built solo in a 48-hour hackathon window, with a deliberate focus on depth over breadth — prioritizing genuine AI organization and retrieval quality over shallow coverage of every possible feature.

4 media files · github.comView

This is Madhav’s work on Wooble.

Build a profile that shows what you can do — and share it anywhere.

Build yours