C

Chirstopher Hanson

About

Detail

Texas, United States

Timeline


work
Job
folder
Project

Résumé


Jobs verified_user 0% verified
  • S&P Global
    Senior Software Engineer
    S&P Global
    Jan 2022 - Current (4 years 9 months)
    Led architecture and implementation of a retrieval-augmented generation (RAG) service using Hugging Face Transformers for embeddings, FAISS for nearest-neighbor retrieval, and Flask based microservices for orchestration, reducing average answer latency from 1.8s to 420ms while increasing answer relevance by 28% as measured by a human evaluation cohort. Architected and deployed PyTorch-based fine-tuning pipelines on AWS SageMaker and EC2 spot clusters with DDP and mixed-precision, cutting fine-tune turnaround from 72 hours to 16 hours and lowering training cost by 42% through automated resource scaling and checkpoint reuse. Designed end-to-end NLP preprocessing workflows using spaCy, custom tokenizers and deterministic text-cleaning pipeline
  • Walmart
    Senior Software Engineer
    Walmart
    Jan 2020 - Dec 2021 (2 years)
    Designed and delivered a document ingestion and embedding pipeline for internal knowledge search using TensorFlow-based encoders and Hugging Face transformers, ingesting 10M+ documents and enabling semantic search across catalogs and support tickets. Implemented microservices in Python and Flask for embedding generation and metadata enrichment, wrapped in Docker and deployed on Kubernetes with Horizontal Pod Autoscalers to manage bursty ingestion traffic during peak catalog updates. Refactored SQL data models (PostgreSQL) and moved large semi-structured payloads to MongoDB with TTL indexes, reducing OLTP contention and improving query latency by 48% for the storefront analytics dashboards. Built a scalable batch inference framework leveragi
  • Cognizant
    Associate / Programming Analyst
    Cognizant
    Jun 2015 - Dec 2019 (4 years 7 months)
    Built and maintained enterprise web applications using JavaScript (ES6), Node.js and React for internal client portals, focusing on modular component design and accessible UI patterns. Developed ETL and SQL scripts to normalize and move large transactional datasets into PostgreSQL and Redshift warehouses, improving reporting latency and enabling near-real-time analytics for stakeholders. Implemented RESTful APIs and microservices in Python and Express, enforcing OpenAPI schemas and improving API test coverage and reliability across integrations. Optimized server-side data access patterns by introducing Redis caching and appropriate TTL strategies which reduced database read pressure and improved average response times by 36%. Created automa
  • F
    Full Stack Developer
Education verified_user 0% verified
  • Golden Gate University
    Bachelor of Science
    Golden Gate University
Projects (professional or personal) verified_user 0% verified
  • R
    RAG-powered Document QA Platform
    Jun 2023 - Current (3 years 4 months)
    Designed and implemented a retrieval-augmented generation platform to answer enterprise document queries by combining dense embeddings (Hugging Face Transformers) and sparse metadata filtering (Elasticsearch). The system ingests PDFs and HTML, extracts structured metadata, computes embeddings with a PyTorch encoder and stores vectors in FAISS for fast similarity search; a Flask API composes the top-k retrieved contexts into prompts for an LLM which runs on SageMaker endpoints. The architecture prioritizes low-latency by integrating Redis caching for hot queries, batched/async inference workers, and autoscaled Kubernetes services for throughput; reproducibility is enforced via DVC for datasets and model artifacts. Implemented a batch ingesti
  • E
    Enterprise NLP Pipeline for NER & Relation Extraction
    Jan 2021 - Dec 2022 (2 years)
    Led development of a production-grade NLP pipeline focusing on Named Entity Recognition and relationship extraction for domain-specific documents. The pipeline combines spaCy preprocessing, custom rule-based NER for edge cases, and transformer-based fine-tuning in PyTorch with label smoothing and data augmentation to improve robustness. Training and evaluation workflows run on managed GPU clusters and use MLflow/DVC for experiment tracking and reproducibility; inference is served via scalable Flask endpoints with