aishaaniagarwal

I built an AI platform that watches you dance and teaches you to dance better — and shipped it live at dancedecoded.in.

• Integrated real-time motion capture using MediaPipe pose estimation, running live in the browser during practice sessions.
• Designed and implemented a custom motion-analysis algorithm that compares a learner's joint positions against reference choreography and scores precision.
• Built the full product end-to-end in Django and Python — the CV pipeline, the feedback UI, and deployment.

A full-stack system that generates complete CBSE question papers and then grades scanned handwritten answer sheets — the whole exam cycle, automated.

• Engineered a document-processing pipeline: PDF text extraction → Ghostscript image conversion → Claude Vision reading actual handwriting → Puppeteer rendering final reports.
• Built the REST API backend in Express handling uploads, AI inference, PDF generation, and async client-server flows.
• Implemented structured prompt engineering and response parsing to turn raw model output into color-coded grading reports.
• Designed class-level analytics — rankings, averages, and the most frequently missed questions across a whole class.

Point a camera at colored physical objects and they become live collision bodies inside a 2D physics world — a tangible interface with no controller at all.

• Implemented the vision pipeline from scratch in OpenCV.js: HSV color segmentation, contour detection, and oriented bounding-box extraction.
• Wrote coordinate transformation and smoothing algorithms to kill detection jitter while staying real-time.
• Integrated Matter.js so rigid bodies spawn and track detected objects' positions and rotations frame-by-frame.
• Built a runtime control dashboard for live tuning of both CV and physics parameters.

A WiFi-enabled servo system whose web interface is hosted directly on the ESP32 itself — control it from your phone's browser, or just clap.

• Programmed HTTP request handling, PWM motor control, and GPIO management in C++ on the Arduino framework.
• Added a microphone module for sound-triggered control, so a clap drives the motor alongside browser commands.
• Built the responsive control page served from the chip, translating taps into low-latency physical movement.
• Did the unglamorous real part too: firmware flashing, serial debugging, and hardware integration.

An AI mediation platform where people privately submit disputes or run live, voice-driven mediation sessions with an AI mediator in the middle.

• Built a secure client-server architecture in Express keeping API keys server-side while routing AI requests through REST endpoints.
• Integrated speech recognition and synthesis via the Web Speech API for real-time spoken conversation.
• Implemented relationship-aware prompting so the mediator's tone adapts — a roommate dispute reads differently than a business one.
• Designed a multi-screen responsive interface with live conversation workflows and smooth transitions.

A batch system that turns a preschool's Excel records and student photos into print-ready, personalized ID cards — built for an actual school's actual problem.

• Built a FastAPI microservice running U²-Net (a deep-learning model) for automatic background removal on uploaded photos.
• Designed a Canvas-based rendering engine with custom layouts, positioning, transformations, and live preview.
• Implemented Excel parsing and automated photo-to-record matching, plus ZIP export for whole batches in one click.
• Deployed the frontend on GitHub Pages and the ML backend on Hugging Face Spaces.

My longest-running project: a career-guidance web app for students who don't have access to counselors, live at unboundonline.in.

• Designed and built the entire app in Next.js, from information architecture to UI.
• Implemented a custom matching algorithm that suggests scholarships, programs, and opportunities to marginalized students.
• Launched and maintained it publicly — real users, real deployments, real bug reports.