Adam T. Bradley

Insurance-conversant Software Developer and AI Practitioner

Experience

Safety Insurance

| Web Application Developer

November 2021 - Present | Boston, MA

  • Generative AI Solutions Architecture: I led the design and implementation of Safety Insurance's first production GenAI application: a Gemini-powered assistant for the MIS department's project management system. I especially enjoyed the challenge of developing a full-stack ingestion pipeline, including custom Python ETL processes for extracting data from an SQL database and a collection of Markdown files, generating embeddings, and indexing the data into Typesense for hybrid vector-semantic search. In response to usage data and comments from colleagues, I migrated the system from a Cache-Augmented Generation (CAG) approach to an agentic tool-use framework to enhance retrieval precision and system scalability. The resulting system saves time for about 50 IT-department employees, eliminating hours of search time and improving efficiency and work satisfaction.
  • Model Optimization & Fine-Tuning: I'm currently exploring the possibility of transitioning from proprietary LLMs to open-weight models by fine-tuning Gemma 4. Leveraging Google Cloud infrastructure (VMs/Compute Engine), I hope to optimize these lightweight models for domain-specific performance, reducing reliance on external APIs while maintaining service quality.
  • Technical Leadership: I served as the primary technical lead for the PMIS assistant initiative, translating business requirements from the MIS department into a technical roadmap, managing the development lifecycle from whiteboard prototyping to production deployment.
  • AI-Powered Agent Assistant: Initially as a graduate school project, I designed an AI assistant and integrated it into the company's insurance producer-facing portal to help insurance agents navigate this tool. This prototype is still under evaluation as we explore wider uses for GenAI technology.
  • Mentoring: I've appreciated opportunities at Safety to help business analysts and junior programmers develop their coding skills. There is an interesting challenge in helping someone translate a business intuition into usable code, and I've enjoyed guiding them through that transition.

Brown University

| Library Programmer

October 2010 - September 2021 | Providence, RI

Developed web applications using Python, PHP, JavaScript, and XSLT.

  • Furnace and Fugue: I served as main developer for the first project of the Library's Digital Publishing Initiative, a digital edition of Michael Maier’s Atalanta fugiens, an esoteric, 17th-century alchemical text. Developed a Python-based static site generator, fugue, using XSL for templating and implemented a client-side search using the Lunr JavaScript library. I also developed style and tool guidelines to provide guidance to development consultants involved in the project.
  • Online Course Reserves Application: Maintaining and improving OCRA, the Library's custom-built course reserves management application, was my primary responsibility for much of my time at Brown. I enhanced OCRA's functionality by integrating free-text and structured API searches, simplifying faculty's addition of course materials. Users praised that improvement as well as the Chrome extension that helped reserve items from across the internet with only one or two clicks.
  • Inscriptions of Israel-Palestine: I improved Solr query efficiency and developed a web-based map interface, furthering the project's research impact and user engagement.

University of Rhode Island

| Part-Time Faculty

August - December 2019 | Kingston, RI

  • I designed and taught CSC 593, "Programming for Scientists", a graduate-level curriculum in introductory computer programming for non-technical professionals.
  • I enjoyed bridging the gap between complex technical concepts and practical application for graduate researchers in the development of data science projects relevant to their diverse academic fields (Engineering, Marine Sciences, English literature).
  • Many student projects resulted in the development of actionable research tools.

Education

Dakota State University

| Madison, SD

Master of Science in Artificial Intelligence

(Expected December 2026)

Specialization & Applied Research:

  • I'm focused on the intersection of Large Language Models (LLMs) and practical enterprise deployment. My insurance agent-facing chatbot developed at Safety became a published paper (IEEE) on LLM-based assistants for insurance agents.
  • Coursework so far has included Neural Networks and AI Ethics, with a focus on building secure, interpretable, and scalable AI systems.

University of Rhode Island

| Kingston, RI

Bachelor of Arts in History and German

Skills

  • AI & Data Science: Large Language Model (LLM) Architecture, Model Fine-Tuning, Prompt Engineering, Data Mining, RAG (Retrieval-Augmented Generation), LLM Tool Design, OpenAI-compatible APIs, LangChain, MCP.
  • Infrastructure: Docker, Git, Github Actions, Cloud Native Architecture (GCP Cloud Run/Storage, AWS S3/Lightsail, Azure AI Foundry), CI/CD Automation.
  • Technical Proficiencies: Python (Django, FastAPI, Numpy, Pandas, PyTorch, TensorFlow, SciPy), JavaScript, API Integration, SQL, MongoDB, Vector Databases (Typesense), Web Application Architecture, CompTIA Security+.
  • Domain Knowledge: Insurance Data Analytics (Associate in Insurance Data Analytics), Cybersecurity Risk Mitigation.
  • Leadership & Management: Project Management Professional (PMP), Agile Methodology, Strategic Planning, Mentoring Non-Technical Stakeholders.

Publications

Introducing Axlerod: An LLM-Based Chatbot for Assisting Independent Insurance Agents

Bradley, Adam and Hastings, John and Ahmed, Khandaker Mamun, 2025 Cyber Awareness and Research Symposium (CARS), 2025