Open source
Research code, end-to-end ML systems, and tools I've built — 22 projects spanning LLM backdoor detection, NLP and speech, healthcare AI, networks, and software engineering.
A “neuro-statistical” watchdog that flags backdoored LoRA adapters without knowing the trigger, using the suspect model’s confidence and semantic-consistency signals and a multi-head verdict: clean, suspicious, or backdoor detected.
Extends the BAIT (S&P 2025) LLM backdoor scanner with sparsemax candidate selection and a controlled zoo of benign, standard, and evasive backdoored LoRA models to locate exactly where the scanner fails while the backdoor still fires.
A backdoor detection system for LLMs based on attention analysis, introducing a “self-centeredness” score to catch sophisticated backdoors that evade confidence-only detectors.
Trustworthy AI for dental diagnostics: per-tooth radiograph analysis designed to support, not replace, a clinician’s judgment. Python, PyTorch, computer vision, FastAPI and React.
An end-to-end PyTorch reproduction of Devign (NeurIPS 2019) on the authors’ released data — composite AST/CFG/DFG graphs, gated graph layers and the Conv readout — plus a commit-disjoint leakage audit of the benchmark.
Implementation of mobility-aware MEC planning with GNN-based graph partitioning (IEEE TNSM 2024): assigning 5G base stations to edge servers to minimise handover cost under load-balancing constraints.
A laptop-friendly lab that fine-tunes a small backdoored model on an 8 GB GPU, then runs five research-paper detection ideas side by side and compares their signals on clean versus triggered inputs.
Attention-ratio detectors for trigger-activated backdoors and contextual hallucinations on Gemma 2-2B and Granite 3.1-3B, with LoRA backdoor training, logistic-regression probes, and attention visualisations.
Trains backdoored Vision Transformers, detects triggered inputs with a linear probe over attention maps, and localises the trigger in the image with a sliding-window search.
The model core of a clinical decision-support system for referable diabetic-retinopathy screening in low-resource clinics, using low-cost fundus cameras and on-premise inference with a human in the loop.
The Next.js web front end for DentalMind.
A Persian mental-health chatbot that detects signs of psychological distress and offers support — published at AbjadNLP 2025 with collaborators at IUST.
Fine-tuned Wav2Vec2 on the ShEMO dataset, combining audio features with transcripts for robust emotion classification in Persian speech.
Sentiment prediction for Persian Instagram posts by fusing VGG-16 image features with a bidirectional LSTM over the text.
Bug-fix work on DadmaTools, the open-source Persian NLP toolkit.
An Android app that collects cellular signal data and estimates base-station locations from RSSI and GPS using circular lateration (Mobile Networks course, IUST).
A Docker-based network-security simulation with attacker, victim, and web-server stations: port scanning, credential testing, and data collection through a Django API.
Assignments and a final project on object detection.
Exercises from the undergraduate Artificial Intelligence course.
Course project for the undergraduate Operating Systems course.
An online shop built with Blazor and deployed on Heroku.
Course project for Data Structures.
No projects match your search.
A generalist biomedical vision-language model distilled from nine biomedical CLIP models.
Persian NLP toolkit developed by Dadmatech.
Fork of antlr/grammars-v4
Grammars written for ANTLR v4.
