Summary

AI/ML Engineer designing and building systems that take models from research into production. Background in linguistics shaped a focus on NLP, LLMs, and RAG. Experience spotting where standard methods break on edge cases — fixing tokenizer vocabulary gaps for crypto terminology, or introducing a new scoring metric to separate agent from patient roles in slur reclamation — and building the fix rather than working around it. Worked across industry and academic settings in Italy, Germany, and Iran.

Work Experience

• NLP/ML Engineer Intern

  2025-08 - 2026-02
  CDC Cartel Damage Claims

  • Robust Data Pipeline: Developed an asynchronous, rate-limited ingestion pipeline with xxhash-based duplicate detection for high-throughput data collection.
  • Domain-Specific Fine-Tuning: Fine-tuned FinBERT for cryptocurrency market sentiment analysis; conducted out-of-vocabulary token analysis to identify sub-tokenisation inefficiencies and optimise model vocabulary.
  • Multi-LLM Consensus Annotation: Engineered a synthetic data annotation pipeline leveraging a multi-LLM ensemble (GPT-4o, Gemini, Mistral) to mitigate single-model bias and improve label quality.
• Junior ML Engineer

  2020-05 - 2023-03
  Digikala Group

  Largest e-commerce platform in MENA.

  • Search relevance: Built Persian NLP components for semantic matching and misspelling correction, helping cut zero-result queries by 7%.
  • Demand forecasting: Built components of XGBoost/Prophet forecasting pipelines that handled 4x Black Friday traffic loads, helping reduce logistics bottlenecks by 10%.
  • Flash-sale pricing: Built automation logic for candidate selection and discount depth in 'Shegeftane' flash sales, supporting ~95% sell-through without eroding margins.
  • Recommendations: Built a module of a hybrid recommendation engine (Association Rules + Collaborative Filtering) for 'Frequently Bought Together,' supporting a 5% lift in Average Order Value via cross-selling.

Education

• Language Technology & Digital Humanities (NLP)

  2026-06
  University of Turin

  GPA: 28.8/30

  Courses: Mechanistic interpretability, RAG systems, Multilingual NLP
• Professional Development Programme in Computer Science

  2021
  Sharif University of Technology

  Courses: Software systems, Algorithms, Data structures, Distributed systems, Statistical learning
• Linguistics

  2019
  Azad University of Hamedan

  Courses: Phonology, Morphology, Syntax, Semantics, Corpus linguistics

Skills

Publications

• Identity, Toxicity, or Complexity? A Language-Specific Feature Selection Approach to Reclamatory Intent Detection

  2026
  Proceedings of EVALITA 2026 (Task A: MultiPRIDE), Bari, Italy

  1st Place (Italian Task): Developed a Hybrid Fusion architecture combining BERT embeddings with engineered sociolinguistic features to detect reclaimed slurs; achieved SOTA F1 of 0.8981 by modeling language-specific syntactic patterns.
• Parametric Stubbornness: Mechanistically Isolating the Layer Shift and Sparsity Gradient of RAG Knowledge Conflicts in Llama-3

  2026
  ACL Student Research Workshop 2026

  Two-phase activation patching on Meta-Llama-3-8B across 452 minimal-pair conflicts; introduces the Sparsity Gradient and Contextual Contamination phenomena in RAG knowledge conflict settings.

Languages

• English

  Full professional proficiency
• Persian

  Native
• German

  Limited working proficiency
• Italian

  Elementary proficiency