01NLP
01
Natural language processing
Sentiment and language understanding for low-resource settings like Persian, pairing transformer context with interpretable lexicon signals.
TransformersParsBERTSentiment
AI / Data Science / ML Engineering / Generative AI
I'm an AI researcher and data scientist building machine learning systems for language, medical imaging, and document retrieval.
M.Sc. AI
4.0 / 4.0 GPA
Research
NLP, CV, RAG
Spectrum
Classical ML to LLMs and Gen AI
Soroush Jaberi
AI researcher / data scientist
Karaj
Iran
Iran
01About
Research-first, build-minded.
I am an M.Sc. Artificial Intelligence candidate and data scientist focused on turning careful research into practical machine learning systems.
My work spans Persian sentiment analysis, medical image segmentation, retrieval-augmented generation, and applied computer vision. I care about experiments that are clean enough to reproduce and products that are understandable enough to trust.
4.0 / 4.0
M.Sc. AI GPA
Autumn 2026
Expected thesis defense
NLP / CV / RAG
Research areas
Working pattern
Research question, baseline, evaluation, iteration, readable implementation.
Principles
- Frame the research question before the model choice.
- Prefer measurable experiments over impressive demos.
- Build readable systems that can be trusted, reviewed, and reused.
02Focus areas
working systems.
Four areas I work across — from the first experiment to something that actually runs.
02MEDICAL
02
Medical AI
Biomedical image segmentation for CT scans, with care for class imbalance, reproducibility, and clinical evaluation.
UNetCT imagingPyTorch
03RAG
03
Retrieval-augmented generation
Document-grounded question answering with vector search and source-aware generation that keeps answers traceable.
LangChainVector searchChromaDB
04VISION
04
Applied computer vision
Real-time pose estimation and image-processing prototypes that turn camera input into useful interaction.
OpenCVMediaPipeReal-time
03Academic
Education, teaching & research.
Graduate AI work, teaching assistant experience, and active research writing.
M.Sc. Artificial Intelligence
Karaj Islamic Azad University
- Overall GPA: 4.0 / 4.0
- Thesis defense expected Autumn 2026
- Machine Learning, Deep Learning, Image Processing
B.Sc. Computer Engineering
Karaj Islamic Azad University
- Overall GPA: 3.74 / 4.0
- Last two years GPA: 3.81 / 4.0
- AI, Data Mining, Algorithms, Computer Graphics
Teaching
2023 - 2025
Teaching Assistant - Machine Learning
Supported graduate students in machine learning workflows and model evaluation, and led Q&A sessions bridging theory and implementation.
2023 - 2025
Teaching Assistant - Data Mining
Designed and evaluated assignments on preprocessing, feature selection, classification, clustering, and pattern discovery.
2022 - 2023
Teaching Assistant - Algorithm Design
Built and assessed problem sets on efficient algorithm design and rigorous complexity analysis.
Enhancing Sentiment Analysis via Ensemble Methods - BERT + VADER
A hybrid sentiment analysis framework combining transformer-based contextual embeddings with lexicon-based polarity signals.
Systematic Mapping Study on Deep Learning-Based Biomedical Image Segmentation Techniques
A systematic study of deep learning architectures across biomedical imaging modalities, evaluation metrics, 3D modeling challenges, and clinical relevance.
04Skills
The tools behind the work.
Grouped by what they're for — modeling, language, vision, and the engineering that ties them together.
Modeling
01Training, evaluating, and iterating on classical and deep learning models.
PythonPyTorchTensorFlowscikit-learnNumPypandasXGBoost
Language
02NLP and document-grounded generation for sentiment, retrieval, and QA.
TransformersBERTParsBERTRAGLangChainChromaDB
Vision
03Medical segmentation and real-time computer-vision prototypes.
OpenCVUNetImage ProcessingPose EstimationMediaPipe
Engineering
04Turning research into readable, maintainable code.
Git/GitHubLinuxJupyterSQLJavaJavaScript
Stack in motion
- Python
- PyTorch
- Transformers
- ParsBERT
- Medical imaging
- Segmentation
- OpenCV
- MediaPipe
- LangChain
- ChromaDB
- Retrieval-augmented generation
- scikit-learn
- NumPy
- pandas
- Computer vision
- NLP
05Selected Work
case studies.
What each project solves, how it was built, and why it matters.
Case 01NLP
Case 01NLP / Deep Learning / Research
Hybrid Persian Sentiment Analysis
- Problem
- Persian sentiment is difficult because informal text, scarce labels, negation, and sarcasm weaken simple polarity models.
- Approach
- ParsBERT contextual embeddings fused with lexicon polarity, combined through ensemble methods and evaluation-led iteration.
- Outcome
- A more interpretable sentiment classification pipeline and the basis of a paper currently under review.
PythonParsBERTVADEREnsembleEvaluation
Case 02RAG
Case 02Generative AI / RAG / Vector Search
LangChain LLM Retrieval QA
- Problem
- LLMs can answer fluently while ignoring private documents or inventing unsupported details.
- Approach
- Document embeddings, ChromaDB retrieval, and grounded generation orchestrated through LangChain.
- Outcome
- A document-grounded QA workflow where answers remain traceable to retrieved source context.
LangChainInstructor-XLChromaDBDolly v2RAG
Case 03MEDICAL
Case 03Medical AI / Computer Vision
Medical Tumor Segmentation
- Problem
- Manual CT tumor delineation is slow, and tiny tumor regions create severe foreground-background imbalance.
- Approach
- PyTorch UNet pipelines for liver and lung tumors with reproducible training structure and image-processing support.
- Outcome
- Segmentation projects across modalities and a foundation for a deep learning biomedical segmentation mapping study.
PyTorchUNetOpenCVMedical ImagingSegmentation
Case 04VISION
Case 04Computer Vision / Real-time
Applied Vision Prototypes
- Problem
- Turning model output into a real-time tool means handling latency, noisy keypoints, and interaction feedback.
- Approach
- Pose-based rep counting with MediaPipe, motion detection, and image-processing prototypes in Python.
- Outcome
- Working camera-input demos including the Live Bicep Counter and related visual interaction experiments.
OpenCVMediaPipePythonPose Estimation
06Timeline
The path so far.
20192026
Swipe through the years
A.01 / 05
2019
Computer engineering foundation
Started building the technical foundation in algorithms, programming, artificial intelligence, data mining, and software engineering.
A.02 / 05
2022
Research & network internship
Worked with Tehran Telecommunication Company on network connectivity data, traffic monitoring, and cloud/networking research.
A.03 / 05
2023
Started M.Sc. in Artificial Intelligence
Began graduate studies with a strong academic focus on machine learning, deep learning, image processing, and applied AI systems.
A.04 / 05
2025
Graduate teaching and research focus
Continued teaching assistant work while deepening research across NLP, medical AI, and retrieval-augmented language systems.
A.05 / 05
2026
M.Sc. thesis defense preparation
Preparing to defend my thesis on improving sentiment classification in Persian texts using transformer-based and lexicon-based methods.
07Contact
Let's build useful AI systems.
Open to research collaborations, AI / data science roles, and projects across NLP, medical AI, retrieval systems, and applied machine learning.
ResearchAI rolesML systems