Sushaanth Srinivasan

AI Browser Extension to Search Websites

engineered a graph-backed crawler and vectorized retrieval to deliver instant, multi-site search from a side panel

Vibe Image Editing Tool

text-driven image editing that fuses GroundingDINO, SAM, and Stable Diffusion with LoRA finetuning for vocabulary-aware inpainting

AI Journal for Devs

led a 9-person team to build an AI-first dev journal with semantic search, code snippet management, and CI-backed workflows

PM Assistant

converted meeting audio into concise emails, auto-created Jira tickets, and generated context-aware boilerplate code

F1 Driver Performance Analysis

analyzed F1 downshift telemetry to map braking behavior to track turns and visualize distance-based performance patterns

System Control Tool Using Gesture Recognition

gesture-driven system control with LSTM/ANN-based face recognition to execute secure commands — hackathon 2nd place

Self-driving Vehicle Simulation

developed an autonomous navigation agent in PyBullet with CV-based path planning and 90% obstacle detection accuracy

Emotion Recognition using Audio Sentiment Analysis

built an audio sentiment engine with ConvNeXt and classical ML to infer musical mood and drive personalized playlist recommendations

Glaucoma Prediction Webapp for Mass Screening

deployed an ensemble U-Net/VGG19/Siamese pipeline for glaucoma screening with a webapp, delivering 99.7% accuracy and 0.85 F1

Multi-lingual Handwritten Character Recognition

researched Siamese and Triplet networks for multilingual OCR, achieving 90% recognition accuracy with bounding-box outputs

Tuberculosis Cavern Classification using 3D-CNN

designed a 3D-CNN for TB cavern classification from CT scans, placing 3rd in ImageCLEF2022 with 75% accuracy

American Sign Language Recognition

created an Android ASL recognizer using Mediapipe and LSTM, deployed with TensorFlow Lite and achieved ~80% accuracy

Physical Networking Assistant

built a networking assistant that records conversations, auto-transcribes, schedules follow-ups, and drafts personalized outreach

Clinical Patient Record Management

built a cross-platform clinical app with CRUD APIs and automated reporting to streamline patient and inventory management for a nonprofit

Los Angeles Crime Analysis

built heatmaps and probabilistic crime estimators using Markov and Chebyshev methods to surface spatiotemporal crime patterns