AI & Machine Learning Engineer | Python Developer
Building reproducible AI systems and practical software across speech, computer vision, NLP, generative models, and developer tooling.
I'm a Python-focused AI and software engineer interested in turning machine learning ideas into reliable, usable applications. My work spans deep learning, computer vision, natural language processing, speech and audio AI, generative models, and AI application engineering.
I care about more than model demos: I focus on data quality, reproducible experiments, explicit evaluation, modular architecture, and maintainable tools. I enjoy working across the full path from dataset inspection and model experimentation to APIs, desktop applications, and developer workflows.
An open research and engineering project exploring whether controlled synthetic defect generation can improve industrial visual anomaly detection.
Implemented foundation: a modular PyTorch package scaffold and CLI, dataset integrity auditing, a VisA one-class dataset adapter, official split validation, and source-image leakage checks.
In progress: generative baselines (DCGAN, conditional GAN, ACGAN, and WGAN-GP), mask-conditioned defect synthesis, and rigorous real-only versus synthetic-augmented evaluation. The project is intentionally evidence-driven: synthetic data is treated as an experiment, not an assumed improvement.
Topics: Generative Models · GANs · Industrial Anomaly Detection · PyTorch · Dataset Engineering
A local-first desktop application and Python CLI for preparing speaker-labelled audio datasets from podcasts, interviews, and other multi-speaker recordings.
The workflow combines audio ingestion, FFmpeg conversion, BS-RoFormer vocal separation, and pyannote speaker diarization. It exports speaker-grouped WAV clips, RTTM, JSON, and source manifests, with a native PySide6/Qt Quick interface.
Status: v0.1.0 early preview. Full model and platform qualification is still in progress.
Topics: Speech AI · Speaker Diarization · Audio Separation · PySide6 · Python
My fork of the open-source Graphify project. I added an optional interactive 3D WebGL view to its generated HTML graph while retaining the existing 2D view and graph-navigation workflow.
Browser-level verification for the 2D/3D interaction is being developed separately; see the repository's pull requests for its current status.
Topics: Open Source · Knowledge Graphs · WebGL · Graph Visualization · Python
A data-feasibility and local-extraction study for evidence-grounded analysis of Persian customer reviews. The project audits real dataset structure and relationships, prepares reproducible benchmark data, and validates extracted aspect/sentiment evidence spans.
The local pilot measures extraction and structural validity; it does not claim verified SnappFood ABSA accuracy. Model-quality evaluation remains a separate milestone.
Topics: NLP · Persian NLP · Aspect-Based Sentiment Analysis · Data Quality · LLM Evaluation
| Area | Tools and interests |
|---|---|
| Machine learning & deep learning | PyTorch, TensorFlow/Keras, scikit-learn, model evaluation |
| Generative AI | GANs, Hugging Face Transformers, LLM application patterns |
| Computer vision & imaging | OpenCV, image processing, OCR, VTK, ITK |
| NLP, LLMs & RAG | LangChain, LangGraph, retrieval pipelines, FAISS, Chroma, Pinecone |
| Speech & audio AI | Speaker diarization, vocal separation, audio processing, dataset preparation |
| Software engineering | Python, FastAPI, PySide6/Qt/QML, Docker, GitHub Actions, Linux |
| Scientific computing | NumPy, pandas, MATLAB, Simulink |
- Reproducibility: explicit configurations, deterministic data splits, and traceable experiment outputs.
- Evaluation before claims: distinguish pipeline validity from model accuracy and report limitations clearly.
- Data integrity: prevent leakage, preserve provenance, and validate inputs instead of silently accepting malformed data.
- Modularity: keep model providers, data sources, and application components replaceable.
- Practical usability: build tools that developers and researchers can actually run, inspect, and extend.
I enjoy building practical AI tools, improving developer workflows, and contributing to open-source projects. Explore my repositories for more work in machine learning, Python engineering, and applied AI.
For collaboration or questions about my projects, connect with me on LinkedIn or send me an email.