Hi, I am Dipanshu Singh

Ideas Den

A collaborative platform to share project ideas, receive feedback, and find global teammates with aligned skills.

Agentic AI for Multimodal Tabular Data Extraction

Built an AI system to detect, extract, and structure multimodal tabular web data while preserving chart and image relationships.

Achieved 91% structure preservation accuracy and reduced manual extraction from 3 hours to under 5 minutes.

VSLM - Very Small Language Model

Designed an end-to-end transformer language model training pipeline using Urban Dictionary data.

AI Flash Cards

Converts raw text and PDFs into personalized, spaced-repetition flashcards using Gemini API.

Crime Analysis: Identifying Risk Areas

Engineered a geospatial GNN over ZIP-code nodes to predict crime hotspots with 86.96% test accuracy.

Tic-Tac-Toe Game

Mobile game with Minimax and alpha-beta pruning, difficulty levels, local storage, and Bluetooth multiplayer.

Context Monitoring App

Context-aware Android app using smartphone sensors and RoomDB for ongoing health monitoring.

Probabilistic Image Inpainting

Analyzed LaMa inpainting under different masks and noise conditions using PSNR, SSIM, and MSE.

Improving LLM Commonsense via Good Listeners

Fine-tuning on good-listener conversation data improved SocialIQA performance for emotional commonsense reasoning.

Social Media Photo Sharing App

Built backend features for uploads, likes, SQL integration, and recommendation-aware content flow.

Image Noise Generator

A web tool to inject noise into images for deep learning experimentation and analysis.