Daniel – Python, AWS, LLM

Deployed an LLM-based phishing detection ranking system using AWS Bedrock, enabling Sales Engineers to identify high-value threat cases; processes hundreds of detections daily with structured, reliable LLM outputs

• Deployed an LLM service through AWS Bedrock;
• Written an API as a wrapper to that LLM so it can be queried/answered;
• Implemented a RAG system that queries a client-side API vector db in Elastic Search to find the necessary record;
• Wrote another client-side API that maintains the vector db - deleted, inserts, updates records - as well as answers to query requests.

Designed and deployed NLP systems for email security, including a few-shot topic classifier (SetFit + contrastive learning) detecting phishing/BEC without traditional indicators and an anomaly detection pipeline for sensitive data leaks and policy violations, serving millions of inferences daily via AWS SageMaker.

• Translated business requirements into project and technological solutions;
• Successfully conducted a thorough investigation on the dataset and what solution is for our use case;
• Model research and validation;
• Model deployment and service (through an API);
• Model monitoring and alerting.

Built an impersonation detection system for email security that prevented any lookalike attacks for VIP people and known contacts.

• Built and maintained Big Data pipelines for data collection and transformation;
• Developed and engineered features for machine learning models;
• Conducted model research, experimentation, and training;
• Designed and implemented APIs for model serving;
• Deployed, monitored, and maintained ML models in production.

Built a semantic editor recommendation system that matched 100+ daily paper submissions with 15,000+ editors using dense embeddings and vector similarity search (early RAG-style retrieval), combining semantic similarity with editor availability and responsiveness for sub-second ranking.

• Built an embeddings-based retrieval system;
• Developed a behavioral model to support the first system;
• Matched 100+ daily paper submissions to a pool of 15,000+ editors;
• Designed and implemented APIs for query processing and indexing;
• Deployed, monitored, and maintained the system in production.