Posts by Collection

portfolio

HAIL: Hardware Automation through Intelligent Language Models

A budget-aware LLM agent workflow that repairs and optimizes AMD Vitis HLS C++ kernels using C simulation, synthesis, and C/RTL co-simulation as feedback.

FPGA-Accelerated Network Intrusion Detection

A small MLP trained on UNSW-NB15, quantized to INT8 and run on a custom FPGA neural-network accelerator for network intrusion detection.

Citexis

A hybrid academic paper discovery platform combining semantic search, keyword search, and recommendations.

AI-Powered Landslide Region Mapping

Semantic segmentation of landslide regions in satellite and UAV imagery to support disaster management.

Agentic Researcher

A CrewAI multi-agent system that researches, writes, and edits a one-page scientific document on any topic.

RAG Researcher

A fully local retrieval-augmented generation (RAG) assistant for analyzing and chatting with research papers.

Near-Earth Object Hazard Prediction

A machine learning classifier that predicts whether a Near-Earth Object is potentially hazardous, with explainable predictions.

Predicting Dengue Fever Outbreaks in Sri Lanka

An Omdena project using time-series forecasting and machine learning to give early warning of dengue outbreaks in Sri Lanka.

MediLeaf

An Android app that identifies medicinal plants and their properties by scanning a leaf, powered by a CNN with over 95% accuracy.

Lucy

An NLP-powered chatbot that answers questions about Nepal Engineering College with around 90% accuracy.

Freyja

An LSTM-based automated essay grading web app, and education category winner at Quantum Hack 2020.

Driver Drowsiness Detection

An end-to-end deep learning application that detects whether a driver is awake or drowsy, with 96% accuracy.

publications

Bridging the Gap in College Information Access Through Natural Language Processing Powered Lucy Chatbot

Published in BOHR International Journal of Smart Computing and Information Technology, 2024

Lucy is an NLP-powered chatbot that answers questions about Nepal Engineering College, making college information easier to access.

Recommended citation: Shah, K. B., Gelal, R., Karki, S., & Shrestha, S. (2024). "Bridging the Gap in College Information Access Through Natural Language Processing Powered Lucy Chatbot." BOHR International Journal of Smart Computing and Information Technology, 5(1), 1-11.
Download Paper

Strict Left Bundle Branch Block Diagnose Through Explainable Artificial Intelligence

Published in IWINAC 2024, Lecture Notes in Computer Science, vol 14675. Springer, Cham, 2024

An explainable machine learning model for detecting Strict Left Bundle Branch Block, developed during my research internship at UPCT.

Recommended citation: Macas Ordóñez, B. C., Garrigós, J., Martínez, J. J., Ferrández, J. M., Karki, S., & Bonomini, M. P. (2024). "Strict Left Bundle Branch Block Diagnose Through Explainable Artificial Intelligence." In Bioinspired Systems for Translational Applications: From Robotics to Social Engineering (IWINAC 2024), Lecture Notes in Computer Science, vol 14675, pp. 504-510. Springer, Cham.
Download Paper

Confocal Microscopy Analysis

Published in Zenodo, 2026

Analysis software for confocal microscopy image data.

Recommended citation: Nemeita, B., Hammann, A. S., Karki, S., & Thoms, S. (2026). Confocal Microscopy Analysis [Software]. Zenodo. doi:10.5281/zenodo.20266295

SMLM Analysis

Published in Zenodo, 2026

Analysis software for single-molecule localization microscopy (SMLM) data.

Recommended citation: Nemeita, B., Hammann, A. S., Karki, S., & Thoms, S. (2026). SMLM Analysis [Software]. Zenodo. doi:10.5281/zenodo.20268284

dSTORMQuant: A Python Package for Post-Processing and Quantitative Analysis of SMLM Datasets

Published in bioRxiv, 2026

An open-source Python package for automated, high-throughput post-processing and quantitative analysis of single-molecule localization microscopy (SMLM) data.

Recommended citation: Karki, S., Nemeita, B., Hammann, A. S., & Thoms, S. (2026). "dSTORMQuant: A Python Package for Post-Processing and Quantitative Analysis of SMLM datasets." bioRxiv. doi:10.64898/2026.06.30.735216.
Download Paper

No Attention, No Problem: DPU-Aware Attention Approximation in Modern YOLO on FPGA

Published in 29th Euromicro Conference on Digital System Design (DSD 2026), IEEE, 2026

A DPU-aware architecture for attention-based YOLO variants (YOLOv26, YOLOv11) on AMD FPGAs, benchmarked across six datasets and all eight DPU configurations on the Xilinx ZCU104.

Recommended citation: Karki, S., Ahmed, Q. A., & Jungeblut, T. (2026). "No Attention, No Problem: DPU-Aware Attention Approximation in Modern YOLO on FPGA." In 2026 29th Euromicro Conference on Digital System Design (DSD), pp. 106-113. IEEE.
Download Paper

FINN-Tro: Exploiting Verification Gaps in Dataflow Inference Accelerators

Published in arXiv, 2026

Introduces FINN-Tro, an attack that exploits a verification gap in the FINN compilation pipeline to insert stealthy hardware Trojans into FPGA dataflow inference accelerators.

Recommended citation: Ahmed, Q. A., Karki, S., & Jungeblut, T. (2026). "FINN-Tro: Exploiting Verification Gaps in Dataflow Inference Accelerators." arXiv preprint arXiv:2609.26824.
Download Paper

FINN4YOLO: Unlocking Real-Time Fisheye Vision with FINN-Accelerated YOLO on FPGA

Published in ResearchGate (poster), 2026

Real-time fisheye object detection with YOLO accelerated on FPGA using the FINN dataflow framework.

Recommended citation: Ahmed, Q. A., Karki, S., & Jungeblut, T. (2026). "FINN4YOLO: Unlocking Real-Time Fisheye Vision with FINN-Accelerated YOLO on FPGA." Poster. doi:10.13140/RG.2.2.15452.14725.
Download Paper

Mitochondrial PEX14 Supports Peroxisome Biogenesis