Tiamiyu Ustaz Hamzat

An ML application that helps the search engine optimization (SEO) team find connections or relations between primary and secondary skills, thereby increasing the relevancy of talent profiles on search engines. For example, we have a primary skill, Python, but finding a connection between it and skills like NumPy, ML, FastAPI, etc., was difficult.

In my team, I developed an ML algorithm that helps find connections between the platform and secondary skills. The result is a skill similarity table with three fields: the main skills, similar skills, and similarity thresholds. This table is saved in BigQuery and updated monthly as retraining is done monthly. It is updated as long as the model performance is not degrading or above a set threshold.

I developed a machine learning solution for exact match detection by leveraging advanced techniques such as sentence transformers to generate embeddings for all database records. The system uses cosine similarity to identify and group similar or duplicate entries. Additionally, it translates natural language queries into SQL code, enabling users to extract relevant or matching records directly from the database.

I fine-tuned the Qwen2.5-7B large language model on a synthetic instruction dataset that I generated from raw texts extracted from PDFs and HTML links referencing multiple PDFs. I evaluated the fine-tuned model using the CyberMetric dataset, a benchmark commonly used to assess the performance of large language models.