Curriculum Vitae
PhD researcher in machine learning for data streams; data scientist building production forecasting systems.
Education
-
2022 — Expected 2027
Ph.D. in Computer Engineering
Bilkent University · Ankara, Türkiye
Advisor: Prof. Fazlı Can · Bilkent Information Retrieval Group (BilIR)
-
2019 — 2022
M.Sc. in Computer Engineering
Bilkent University · Ankara, Türkiye
Thesis: Evolving Text Stream Classification with a Novel Neural Ensemble Architecture · repository
-
2013 — 2018
B.Sc. in Computer Engineering
Iran University of Science and Technology · Tehran, Iran
Thesis: Driver Drowsiness Detection using Raspberry Pi
Experience
-
Sep 2025 – Present
Data Scientist
Enterprise retail AI · Türkiye
- Cut lost sales by 10% for an enterprise retail client by improving demand forecast accuracy and the replenishment decisions driven from it.
- Own demand forecasting and replenishment models in production for three enterprise retail clients, covering roughly 1M SKU-store pairs — from feature pipeline through model release.
- Build and operate large-scale ML pipelines on Databricks, PySpark and Airflow: feature generation, training, backtesting and scheduled inference.
- Design the offline and online evaluation behind model releases — backtesting setup, holdout construction and drift monitoring on live series.
- Deliver inventory-visibility models and next-best-action recommendations for client planners, working with product and engineering to ship them.
-
Apr 2019 – Present
Senior AI/ML Researcher part-time
Bilkent University · Ankara, Türkiye
- Named senior researcher on three government-funded R&D programs (TÜBİTAK 117E870, 120E103, 125E060); lead the ML workstream on classification, anomaly detection and concept-drift adaptation over large-scale streaming and event data.
- Designed AdaNEN, a neural ensemble for evolving data streams, improving classification accuracy by up to 8.8% across 13 benchmark datasets (ACM TKDD).
- Co-designed LACE, an unsupervised concept-drift detector for multi-label streams that cuts detection delay by 63.5% versus the previous best unsupervised method at a 0% missed-detection rate (ACM CIKM 2026).
- Proposed the Adaptimizer optimizer and an On-the-Fly Adaptation (OFA) framework for continual LLM adaptation under temporal drift, evaluated across two decades of news (ACM CIKM 2025); implemented in PyTorch and HuggingFace.
- Own projects end-to-end — dataset construction, training, evaluation metrics, and offline/online experiments — turning research into reusable, well-evaluated models.
-
Feb 2019 – Present
Teaching Assistant part-time
Bilkent University · Ankara, Türkiye
- Teach and mentor students in Information Retrieval, Algorithms and Computer Organization, explaining complex ML concepts to technical and non-technical audiences.
- Supervise student projects on online learning, neural retrieval and ranking systems.
-
Jun 2018 – Sep 2018
Android Developer Intern
Petanux GmbH · Bonn, Germany
- Built an Android app for online price comparison across multiple vendors.
-
Jun 2017 – Oct 2017
Software Engineer
Shams Clinic · Tehran, Iran
- Built a clinic management system for appointment scheduling, staff coordination and financial tracking, used in daily operations.
Publications
-
2026LACE
LACE: Unsupervised Concept Drift Detection in Multi-Label Data Streams Through Label Cluster Evolution
ACM CIKM
An unsupervised concept-drift detector for multi-label streams that cuts detection delay by 63.5% against the previous best unsupervised method, at a 0% missed-detection rate.
DETECTION DELAY −63.5% · 0% MISSED · MULTI-LABEL STREAMS
-
2025LLM-OFA
LLM-OFA: On-the-Fly Adaptation of Large Language Models to Address Temporal Drift Across Two Decades of News
ACM CIKM
An On-the-Fly Adaptation framework and the Adaptimizer optimizer for continually adapting LLMs under temporal drift, evaluated across two decades of news.
2 DECADES OF NEWS · CONTINUAL LLM ADAPTATION
-
2024AdaNEN
A Novel Neural Ensemble Architecture for On-the-Fly Classification of Evolving Text Streams
ACM TKDD
AdaNEN — a neural ensemble for evolving data streams that improves classification accuracy by up to 8.8% across 13 benchmark datasets.
ACCURACY +8.8% · 13 DATASETS · SINGLE-PASS
-
2023BELS
A Broad Ensemble Learning System for Drifting Stream Classification
IEEE Access
A broad-learning ensemble for classification over drifting data streams.
BROAD ENSEMBLE · DRIFTING STREAM CLASSIFICATION
Technical skills
Machine Learning & LLMs
- PyTorch
- HuggingFace Transformers
- Scikit-learn
- NumPy
- pandas
- Deep Learning
- Neural Networks
- Large Language Models
- Fine-tuning (LoRA / PEFT)
- Retrieval-Augmented Generation (RAG)
- Vector Databases
ML Methods
- Time-Series Forecasting & Demand Prediction
- Classification
- Anomaly Detection
- Recommendation
- Ensemble Learning
- Online & Continual Learning
- Concept-Drift Adaptation
- Time-Series Foundation Models
- Feature Engineering
- Experimentation (offline/online)
- Statistical Hypothesis Testing (Friedman, Wilcoxon)
- Model Evaluation
Data & MLOps
- SQL
- Databricks
- PySpark
- Airflow
- FastAPI
- Docker
- Linux
- Git
- CI/CD
- AWS (EC2, S3, Bedrock)
- Azure ML
Programming
- Python (primary)
- SQL
- C++
- Java
- C#
Awards & scholarships
- 2010
- Gold Medal (×2) — Iranian National Student Olympiads — Physics and Research
- 2019–present
- TÜBİTAK Research Scholarships — Funded researcher on three national R&D programs (117E870, 120E103, 125E060)
Professional service
- Conference reviewer
- ACM SIGIR (2023–2025) · ACM CIKM (2023–2026) · SIGIR-AP (2024–2026)
Languages
- Azerbaijani
- Native
- Persian
- Native
- Turkish
- Fluent
- English
- Advanced — TOEFL iBT 104
Contact
- pouyaghahramanian@gmail.com
- Scholar
- Google Scholar
- GitHub
- PouyaGhahramanian
- pouyaghahramanian