SharayuRasal

Problem

High‑volume learning platforms lacked consistent content retrieval, causing slow recomputation and no reliable way to deliver offline personalized content for users with limited connectivity.

Solution

Built a full‑stack AI recommender using Django REST, React, PostgreSQL, and AWS S3, with Celery and Redis handling async ingestion, caching, and aggregation. Added an ML‑driven ranking module using vector embeddings, cosine similarity, and top‑K matching for fast, personalized suggestions.

Insight

Shows how combining classical IR, lightweight ML, and distributed caching enables fast, offline‑friendly, production‑grade personalization with strong reliability and cloud scalability.

Problem

Early detection of depression from longitudinal data is challenging and often under‑instrumented.

Solution

Combined HRV features with a 1D‑CNN and emotion‑aware CNN pipeline to classify depression severity from multimodal inputs.

Insight

Demonstrates how multimodal signals can be fused for more robust mental‑health screening tools.

Problem

Users needed an intuitive way to search through large photo collections using everyday language queries like 'show me photos with dogs' instead of manual tagging.

Solution

Built a serverless architecture using Amazon Lex V2 for natural language processing, Rekognition for automatic image labeling, OpenSearch for full-text search, and S3 for scalable storage. Lambda functions handle photo indexing and search, with API Gateway exposing REST endpoints.

Insight

Demonstrates how to combine multiple AWS AI/ML services into a cohesive, production-ready application with CI/CD automation using CodePipeline and CloudFormation.