PAPER-01 · IEEE ACCESS
Failure Geometry in Object Detection
4D failure vector & mCRS metric across YOLOv8, DETR, RT-DETR on 17 corruptions.
Writing Phase —
Explore →
MTech AI/ML · ADIT, Gujarat · GPA 9.2/10
An AI researcher working on robustness, fairness, and safety across computer vision and NLP systems. Research explores failure modes in object detection, bias in routing architectures, and cross-lingual safety in LLMs — supported by retrieval-augmented generation and systematic evaluation. Committed to building AI that is reliable, auditable, and performant beyond the benchmark.
Click any card to open an interactive, animated architecture breakdown — or download the PDF directly.
4D failure vector & mCRS metric across YOLOv8, DETR, RT-DETR on 17 corruptions.
Routing fairness ≠ representation fairness in MoE. Validated via FDI across 5 datasets.
Mechanistic interpretability: Apertus 8B vs Llama-3 8B in English × Gujarati.
A retrieval-augmented generation pipeline — deployed live and fully open-sourced, end to end.
Professors, collaborators, fellow researchers — I read every transmission.
thakerparth12@gmail.com github.com/partthaker linkedin.com/in/partthaker📑 Download CV
⬇ Download My CV