I'm Jahnavi Singh,
ML & Computer Vision
M.Tech @ IIT Delhi. Building real-time vision systems, generative models, and semantic communication pipelines. I implement algorithms from scratch — from GMMs to neural networks.
IIT Delhi · Research & ML
Cyber Vision — GMM Background Subtraction
Real-time background subtraction using Gaussian Mixture Models (Stauffer & Grimson, 1999). Interactive web demo with live camera feed and tunable parameters (α, K, T, B). Four-panel visualization: original, FG mask, foreground overlay, reconstructed background.
Neural Network from Scratch
Built neural networks using NumPy with backpropagation and mini-batch SGD. Architecture search comparing Sigmoid vs ReLU. Evaluated using Macro-F1 and benchmarked against scikit-learn MLP.
Semantic Communication for Image Generation (ongoing)
Designing a pipeline using GNNs to construct scene graphs from structured inputs. Generating layouts and images with GANs and diffusion models. Exploring end-to-end semantic encoding.
Object Detection & Transformer Modeling
Fine-tuned YOLO for aerial object detection. Implemented FID metric for generative evaluation. Trained transformers with/without pretraining — analyzed performance gains.
Previous Work · Industry & Learning
Trade Vogue — E‑commerce Platform
Full‑stack e‑commerce application built during internship at Venture Pact. Features product catalog, shopping cart, user authentication, order management, and live deployment. Scalable backend with Node.js, MongoDB, Redis.
Architects on Mission — Design Agency Website
Static multi‑page website for a fictional architecture firm. Built during early front‑end learning phase. Features responsive layout, service sections, gallery, testimonials, and contact form.