Zaeem Ansari

Machine Learning Engineer @Allia Health

Reading, GB

Built the entire data model and structure using Python Neomodel, organizing over 12 types of clinical entities and relationships in a Neo4j graph database.– Designed the overall LLM strategy for clinician queries, including fine-tuning, query routing, BM25, andRetrieval-Augmented Generation (RAG), improving retrieval quality and system scalability.– Led the development of AI-driven treatment plans and assessments, improving response accuracy by 30% through optimized prompt engineering.– Structured all response handling by implementing a class-based system, ensuring maintainability and scalability.– Standardized responses in JSON format, streamlining backend integration for seamless data processing.– Built a retrieval-augmented generation (RAG) pipeline using LangChain, leveraging psychological research papers, journals, and articles.– Automated data scraping, cleaning, and storage in AWS S3, using Selenium and BeautifulSoup4, improving document retrieval efficiency by 40%.– Fine-tuned LLaMA 3.1 70B on Hugging Face and initially deployed on EC2 GPU instances, incurring a $2,500/month cost.– Implemented a serverless solution (Fireworks), reducing infrastructure costs to $0 when idle, improving inference efficiency by 60%.– Integrated Amazon Neptune Graph Analytics with LlamaIndex, creating a knowledge graph (KG) extractor and improving query accuracy by 35%.– Developed a PropertyGraph index and optimized response generation by combining deterministic (structured) and non-deterministic (LLM-driven) nodes using OpenCypher.– Benchmarked Neo4j vs. Neptune for accuracy and response relevance to enhance knowledge retrieval.

National University of Computer and Emerging Sciences

Bachelor of Science - BS, Computer Science

Royal Holloway, University of London

Master of Science - MS, Artificial Intelligence