I'm a Computer Science & Engineering student at Pimpri Chinchwad University (CGPA: 8.76), and most of my time outside coursework goes into building and breaking AI models — mostly in Computer Vision and Deep Learning.
I like picking a real, slightly messy problem — a medical imaging dataset, a video feed, a pile of unstructured documents — and seeing if I can actually get a model to work on it, not just in a tutorial sense. A lot of my projects started as "let me see if this is possible" and turned into something I kept iterating on.
Right now I'm spending most of my time on Deep Learning fundamentals, getting deeper into LLMs and Retrieval-Augmented Generation, and working through DSA in Java on the side. I also picked up Japanese a while back — JLPT N5 certified, slowly working toward N4 — mostly because I enjoy learning something new from scratch, the same way I approach ML.
class Riddhi:
def __init__(self):
self.role = "CS Undergrad"
self.interests = ["Computer Vision", "Deep Learning", "LLMs", "RAG"]
self.currently_learning = ["DSA in Java", "RAG pipelines", "Japanese (N5 → N4)"]
def what_i_like(self):
return "Taking a real dataset, building a model, and seeing if it actually works"Machine Learning Deep Learning Computer Vision Large Language Models
Retrieval-Augmented Generation AI for Healthcare Data Science Smart City Systems
Languages
AI / Machine Learning
Data Science
Databases & Tools
Web & App Dev
|
Computer Vision · Medical Imaging Detects Polycystic Ovary Syndrome from transvaginal ultrasound images by segmenting ovarian follicles. Curated a hybrid medical imaging dataset, hand-annotated follicles with polygon segmentation, and expanded the dataset from 308 → 844 images through augmentation. Trained and compared YOLOv8 through YOLOv11. Best model: Precision 87.5% · Recall 84.5% · F1 86.0%
|
Computer Vision · Environmental Monitoring An independent project to flag illegal waste dumping from video/image feeds. Built a custom dataset from scratch — manually labeled 637 images, expanded to 1,723 through augmentation — then trained a YOLOv11s detector. Results: Precision 83.2% · Recall 79.4% · F1 81.3%
|
|
OCR + LLMs + Layout Understanding An end-to-end document processing pipeline combining OCR, layout analysis, and LLM-based information extraction for structured understanding of multilingual documents.
|
Smart City Resource Optimization — 🏆 2nd Runner-Up, National Hackathon A multi-module system covering water leak detection, waste collection optimization, public health monitoring, and risk-based decision support, built end-to-end during a hackathon.
|
AI / Data Science Intern Worked across document AI systems, OCR pipelines, LLM integrations, and Dataiku workflows, building AI-powered automation tools. Responsibilities included research, MVP planning, backend documentation, and AI module development.
- 🥉 2nd Runner-Up — Smart City AI Hackathon (CityBrain AI)
- 🔬 Built a handful of independent Computer Vision projects end-to-end — from dataset curation to trained, evaluated models
- 📊 Worked with real, messy datasets rather than clean, pre-packaged benchmarks
- 🇯🇵 JLPT N5 certified
- Going deeper into LLMs and RAG pipelines
- Strengthening DSA fundamentals in Java
- Picking up a new CV problem or paper to reimplement
- Slowly working toward JLPT N4
If you're working on something in Computer Vision, Deep Learning, or LLMs, or just want to talk through a project, feel free to reach out.
⭐ Thanks for stopping by — feel free to look through the projects or say hi.