Hemesh RM Data Solutions Engineer

Developed a web application that enables users to discover and generate personalized recipes using GroqCloud's LLM (Llama model).

• Integrated AI to generate meal ideas tailored to user preferences
• Supported filters for cuisine, nutrition goals, and allergies
• Stored and retrieved user history with Firebase for seamless UX

Built a semantic recommendation system using embeddings to help users discover relevant books based on content similarity and genre preferences.

• Embedded 7,000+ book descriptions into 384-d vectors using all-MiniLM-L6-v2 for similarity search
• Integrated Hugging Face and LangChain with Qdrant for fast cosine retrieval and metadata filtering
• Deployed a Gradio app with a zero-shot classifier (∼80% accuracy) for fiction vs non-fiction labeling

Designed and trained an AI agent using NEAT to autonomously master the Flappy Bird game through neural network evolution.

• Developed autonomous gameplay using NEAT to evolve neural networks without human input
• Implemented collision detection, game mechanics, and dynamic pipe difficulty for realistic simulation
• Achieved 100% survival rates by optimizing bird movement based on pipe positions and altitude

Built an AI-powered system for real-time football analysis using computer vision techniques to extract insights.

• Used YOLOv8 for object detection of players, referees, and the ball, achieving 79% mAP and 80% IoU
• Clustered players by jersey color using KMeans with a silhouette score of 0.6; added optical flow to stabilize tracking with camera motion
• Computed player speed and distance via perspective transformation for real-world accuracy

Created a Tableau dashboard analyzing Netflix’s global catalog to uncover content strategies and market potential.

• Identified high-demand regions like the U.S. and India, and untapped areas such as Africa and Eastern Europe
• Revealed that 68% of the content consists of movies, highlighting Netflix’s focus on versatile viewing patterns
• Surfaced trends in genres like Documentaries and Dramas, and growth in niche categories such as true crime

Designed a robust ETL architecture for automotive sales data using industry-standard patterns and modern data engineering tools.

• Implemented scalable pipelines with Medallion Architecture, Delta Lake, and Unity Catalog
• Built a Star Schema with Fact and Dimension tables; automated incremental loads via stored procedures in ADF
• Used SCD modeling in Databricks to ensure high data consistency and traceability

Built an AWS-based data pipeline to analyze regional YouTube video trends, leveraging serverless tools for transformation and visualization.

• Ingested and transformed trending video data using AWS S3, Glue, and Lambda into Parquet format
• Performed SQL-based analytics with Athena and created interactive dashboards in QuickSight
• Used AWS Glue Catalog for automated schema detection and metadata management

Built a real-time data pipeline to simulate and process stock market data, leveraging Kafka and AWS analytics services for streaming insights.

• Implemented producer-consumer pipelines in Kafka to ingest and stream real-time stock data
• Simulated data using Python and Boto3, writing outputs to S3 for downstream processing
• Used AWS Glue Crawlers to catalog datasets and Athena for on-demand SQL analysis

Built a sentence-level lip-reading system using deep learning reduce reliance on audio-based speech recognition.

• Built a LipNet-inspired model using STCNNs, RNNs, and CTC loss for sequence prediction
• Trained on the GRID dataset and developed a Streamlit app for real-time lip-reading
• Integrated into an assistive communication system for improved accessibility

Built a web application that converts blog URLs into summarized videos using GPTScript and FFmpeg pipelines.

• Used GPTScript to summarize content and generate visual + audio media
• Implemented Node.js backend to handle FFmpeg workflows and API requests
• Developed a React frontend to present generated videos interactively
• Enhanced content accessibility and engagement through automation

Built a full-featured theatre management platform with real-time seat tracking and secure payment integration.

• Developed role-based portals for Users, Admins, and Theatre Owners with secure access
• Built with Ant Design, Redux Toolkit, JWT + BCrypt authentication, and deployed on Render
• Integrated Stripe for payments and implemented show scheduling, seat selection, and ticket control

Built a deep learning system to classify microscopic algae images for environmental monitoring.

• Developed a deep learning system using CNN, AlexNet, and ViT to classify algae images
• Processed FlowCam DB images with data augmentation, achieving 98% top-5 accuracy
• Deployed at the City of Bloomington’s office, integrating into a live preprocessing-to-display pipeline