Rahul Kashyap
Selected Work

Projects and systems

Open-source tools, research systems, and applied AI, spanning continual learning, computer vision, and generative AI.

clearn-ai

Open Source

Open-source PyTorch library enabling models to learn sequential tasks without catastrophic forgetting. Features model.diff() for retention reports. Achieved 82–88% accuracy on CIFAR-100 where baselines dropped to 8%.

PyTorch · Continual Learning · CIFAR-100

PIT

Open Source

Semantic version control for production LLM prompts. A Git-for-prompts tool with pit bisect for broken version identification, pit replay for time-travel, and a performance-based query language.

Python · LLMs · Version Control

Course On Demand

AI Application

AI platform transforming research papers into structured learning modules with video explanations, coding exercises, and assessment systems.

Generative AI · NLP · Full Stack

AI Bone Fracture Detector

Deep Learning

Multimodal deep learning model combining ResNet-50 for X-ray analysis with LSTM for textual descriptions, improving diagnostic accuracy through complementary modalities.

ResNet-50 · LSTM · Medical AI

Cervical Cancer Detection

Computer Vision

YOLO-NAS classifier for cervical pap-smeared cells in collaboration with biologists. Includes CI/CD pipeline via Django for real-time clinical detection.

YOLO-NAS · Django · CI/CD

Pill Image Classification

Computer Vision

Unsupervised ML pipeline analyzing 8,693 pill images from the National Library of Medicine's Pillbox dataset. Integrated with YOLO-NAS for real-time classification.

YOLO-NAS · Unsupervised ML · Image Segmentation