Production NLP & Computer Vision Engineer

Austin Allen

919-***-**** *********@*****.*** linkedin.com/in/austindallen github.com/adallen93 SUMMARY

AI Engineer with production experience building LLM agent systems, retrieval-augmented generation pipelines, and NLP-driven data pipelines in Python. Skilled in prompt and model work, vector search and embeddings, and evaluating model outputs for accuracy and relevance. Comfortable researching and implementing techniques from recent AI work and collaborating cross-functionally with engineering and clinical teams to ship AI products. EXPERIENCE

AI Engineer DataThink Rexburg, ID (remote) Jul 2025 – Present

• Co-architected an internal MCP server exposing composable LLM agent workflows and tools to Claude Code and LangGraph, moving the team away from monolithic prompting toward scoped, reusable components.

• Led a 3-person team building a RAG pipeline that grounds LLM agents in domain-specific regulatory documentation, taking a raw-data-to-structured-output workflow into production across 4 studies.

• Built an end-to-end PHI redaction pipeline using spaCy NER and rule-based pattern matching, processing roughly 20,000 patient messages with zero PHI leakage detected in review.

• Migrated a production knowledge base to MongoDB Atlas vector search with self-managed AWS Bedrock embeddings, cutting hosting cost about 90% while matching the similarity metric to the embedding model's training objective. Computer Vision Engineer Kitware, Inc. Carrboro, NC Jan 2025 – Jul 2025

• Fine-tuned a ResNet-50 model on roughly 500 whole-slide images, reaching 80 to 90 percent accuracy, precision, and recall across three segmentation classes.

• Built a reporting pipeline integrating model outputs into a clinical visualization tool, iterating directly with pathologist collaborators to match real workflow requirements. Research Assistant Duke University, Dept. of Biostatistics and Bioinformatics Durham, NC Aug 2024 – Dec 2024

• Trained a CNN classifier on a 2.4-million-example genomics benchmark spanning 130 bacterial classes, reaching approximately 96 percent in-distribution accuracy.

• Designed a training and validation paradigm using softmax-based uncertainty thresholds to flag out-of-distribution inputs, reaching high-80s precision and recall on novel-strain detection. EDUCATION

Master of Biostatistics, Data Science Emphasis

Duke University

Durham, NC

Aug 2023 – May 2025

Coursework: Applied Deep Learning (CNNs, Transformers), Computer Vision, Big Data Systems TECHNICAL SKILLS

AI & LLM Engineering: LLM Agents, RAG, LangGraph, MCP, Prompt & Model Tuning, spaCy, NLP, Vector Search & Embeddings Machine Learning: Deep Learning, PyTorch, CNNs, Scikit-learn, Random Forests, UMAP, HDBSCAN Programming: Python, R, SQL

MLOps & Cloud: AWS, AWS Bedrock, GCP, BigQuery, MongoDB Atlas, Docker

Contact this candidate