MoJa Ardestani

• Designing searchable knowledge base for RAG systems aligned with business needs
• Evaluating classification methods to improve metadata quality and enable effective cross-format retrieval
• Collaborating cross-functionally with scientists and product managers to define roadmap milestones
• Investigating feedback mechanisms to continuously improve system performance and usability

• Advised startups on LLM system design, supporting ideation, architecture, and deployment planning
• Provided mentorship on fine-tuning and RAG, adapting recommendations to client constraints
• Guided ethical LLM deployment, including bias mitigation and output safety in real-world use cases
• Explained ML strategies to both technical and non-technical team members to support informed decisions

• Advised on designing an LLM-based pipeline for automated report generation using a RAG framework
• Supported evaluation of factual accuracy, addressing hallucination and explainability
• Recommended strategies for embedding fine-tuning to improve domain-specific retrieval

• Developed and deployed an advanced hybrid book recommendation system for Taaghche
• Combined word embeddings with fine-tuned BERT + CNN, boosting recommendation relevance by ~20%
• Used Siamese networks for pairwise similarity scoring, enhancing cold-start and long-tail ranking
• Applied SVR and KNN as fallback models to handle sparse user data, reducing false negatives by ~15%

• Proposed a recall evaluation pipeline for LLM-generated answers in free-form and list-based QA
• Developed a modular system combining fact extraction, candidate selection, and entailment checking
• Applied the pipeline to high-stakes domains such as law and medicine, emphasizing factual completeness
• Achieved up to +40% F1 improvement over lexical baselines and +30% over holistic prompting

• Studied overfitting mitigation in incremental learning via custom adaptation and weighting strategies
• Proposed dynamic backpropagation weights based on inverse cluster size via incremental clustering
• Designed learning schemes tailored to changing data distributions in long-term tasks
• Improved model generalization in evolving data environments through adaptive optimization

Supporting undergraduate and graduate students in computing science courses, providing guidance on assignments and projects.

• Conducted office hours and tutorial sessions for various computing science courses
• Graded assignments and provided detailed feedback on student work
• Mentored students on research projects and final course projects
• Assisted in developing course materials and practice problems

Teaching machine learning concepts and practical applications to students at the Tehran Institute of Technology.

• Designed and delivered comprehensive machine learning curriculum
• Taught fundamental concepts including supervised and unsupervised learning
• Guided students through hands-on projects and practical implementations
• Provided mentorship on machine learning applications and best practices

Supporting undergraduate students in computer engineering and computing science courses.

• Assisted professors in various computer engineering courses
• Conducted laboratory sessions and provided technical support
• Graded assignments and provided constructive feedback
• Helped students understand complex technical concepts