Harini Gona
AI/ML engineer focused on building and evaluating AI systems
New to Contra
Harini is ready for their next project!
I worked on developing a non-electric air purification product from the early customer-discovery stage through business-model development. I conducted outreach with 100+ potential customers to understand their needs, gathered feedback, and researched the market, competitors, TAM/SAM/SOM, and commercial feasibility. I also worked with technical and business mentors to refine the product strategy, customer value proposition, and venture proposal, including communication with stakeholders at Northeastern University.
1
2
59
I analyzed energy and environmental data across 14 countries over a 20-year period using SQL. I explored trends in fossil fuel consumption, renewable energy usage, and CO₂ emissions, using SQL queries to compare countries and identify significant changes over time. I also created visual summaries to communicate the findings and investigated the relationship between the decline in fossil fuel use, growth in renewable energy, and changes in CO₂ emissions.
1
66
Developed an AI-powered image enhancement system using teacher-student knowledge distillation, achieving 92% accuracy and 3× faster inference on edge devices. Benchmarked model performance, analyzed inference quality and accuracy trade-offs, and evaluated different model configurations to improve efficiency and reliability. Worked with Python, PyTorch, OpenCV, and computer vision techniques to develop and evaluate the system.
1
65
I built an AI-powered apparel identification system designed to identify products even when they don't have a barcode. The core idea is to take an image of an apparel product, detect the relevant visual information, and retrieve the most visually similar products from a database of 10,000+ product images.
I used YOLOv8 for object detection, CLIP for generating visual embeddings, and FAISS for efficient similarity search. I also evaluated the system using precision/recall and Recall@K, and analyzed incorrect matches and edge cases to improve the reliability of the results.
The UI provides a practical way to interact with the system and see the identification/retrieval results rather than just exposing the underlying model. I built the project to explore how computer vision and similarity search can solve a real-world product identification problem where traditional barcode-based identification isn't available.
1
63