Behtash Kiani

• Prepared, cleaned and validated structured and unstructured datasets for production machine learning pipelines, maintaining strict data integrity.
• Streamlined experiment tracking and metadata organization to improve result comparability and ensure end-to-end model reproducibility.
• Supported the core ML engineering team by implementing automated data-processing components using open-source tools to improve workflow throughput.
• Authored comprehensive technical documentation mapping data flows across pipeline stages, reducing cross-team onboarding and execution bottlenecks.

Python / FastAPI / NumPy / Ollama

• Engineered a production-grade RAG support system with a modular multi-file Python architecture and an asynchronous FastAPI backend.
• Built a zero-cost TF-IDF vector retrieval engine from scratch using NumPy and L2-normalized cosine similarity for precise local context search.
• Integrated an offline Llama 3.2 model via Ollama, using strict temperature tuning and data grounding to minimize hallucinations.

Python / UML

• Designed and developed a system for managing university courses using Python and UML class diagrams.
• Implemented features for adding and removing students, managing lecturer roles and retrieving course information.

JavaScript / Node.js / MongoDB / Express

• Built a web application using the Spotify API to display the user’s top songs with album artwork.
• Stored listening history in MongoDB and served content dynamically through a Node.js/Express backend.