Nicholas Smith

M

Senior Al Engineer

MediSearch

Apr 2023 - Current (3 years 5 months)

Built a production-grade AI search engine using multi-agent workflows, RAG pipelines, and a React/FastAPI stack. Developed a retrieval-augmented generation (RAG) pipeline using LangChain and LangGraph, which indexed over 100k+ peer-reviewed medical sources to enhance answer fidelity and ensure evidence-backed responses. Created a multi-agent system in Python that delineated responsibilities among agents for query parsing, vector-based evidence retrieval, and GPT-3 answer synthesis, thereby improving modularity and facilitating future scalability. Fine-tuned GPT-3 and specialized LLM checkpoints on medical Q&A datasets, achieving a 90% score on USMLE-style benchmark exams to validate clinical accuracy. Designed hallucination detection by cro

Al Engineer

Amazon

Mar 2021 - Mar 2023 (2 years 1 month)

Focused on enhancing large-scale recommendation systems for 10M+ users, using AWS AI tools and infrastructure. Developed and deployed BERT-based sequence models using Amazon SageMaker, which improved the relevance of product recommendations by analyzing user behavior patterns. Optimized real-time feature stores by integrating DynamoDB and Redis, reducing latency in personalized recommendations to under 200 milliseconds. Utilized AWS Inferentia and TensorRT to accelerate model inference, achieving a 32% reduction in processing time for recommendation algorithms. Implemented A/B testing frameworks to evaluate multiple model variants across different regions, facilitating data-driven decisions for model improvements. Migrated legacy Java-based

Machine Learning Engineer

AlphaSense

Mar 2016 - Feb 2021 (5 years)

Focused on scaling NLP infrastructure and vector search systems for financial document analysis. Implemented semantic search using BERT, which increased the accuracy of SEC filings queries by 45% through smarter query expansion. Reduced manual labeling by automating the annotation of over 2 million financial documents using Snorkel-based weak supervision. Enabled 35% faster model training by migrating Elasticsearch clusters to Kubernetes and configuring GPU resource auto-scaling. Created a WebSocket-powered API that leverages FinBERT to analyze earnings calls in real time. Built a system combining Elasticsearch's keyword search and FAISS vector similarity to boost the accuracy of retrieving financial documents. Streamlined Apache Spark jobs

Full Stack Developer

DAT Freight & Analytics

Sep 2013 - Mar 2016 (2 years 7 months)

Modernized legacy freight management systems and integrated IoT sensor data for real-time monitoring. Implemented Angular dashboards combined with Mapbox for live monitoring of 10k+ daily shipments. Transformed a monolithic Ruby on Rails application into Dockerized microservices. Created REST APIs using Node.js to integrate IoT sensor data from freight vehicles, enabling real-time tracking and analytics. Enhanced shipment history queries in PostgreSQL to cut report time from 5 minutes to 45 seconds and boost data retrieval speed. Integrated OAuth 2.0 authentication on driver mobile applications, which ensured PCI DSS compliance and enhanced security. Implemented Apache Airflow for ETL pipeline orchestration to streamline workflows and enhan

About

Detail

Timeline

Résumé