Thorin Tabor

He / Him

Thorin Tabor

Staff Software Engineer  ·  Agentic AI, MCP & RAG

UC San Diego & Broad Institute of MIT and Harvard

San Diego, California, United States

thorin@tmtabor.io LinkedIn GitHub

About

I have been coding recreationally since grade school. Professionally, I've spent nineteen years turning that into a career, most recently as a Staff Engineer at UC San Diego, where I build agentic AI systems using RAG, MCP and LLM orchestration for scientific research. Academically, I have a master's degree in computer science.

Along the way I've co-founded three companies, including a Google Startups-backed SaaS, contributed to a range of open source projects and co-authored seven peer-reviewed publications. In recent years my focus has narrowed to the hard parts of applied AI: making agents reliable, keeping hallucinations out of high-stakes domains and building infrastructure that holds up in production.

Recreationally, I'm still at it.

Experience

University of California, San Diego

Nov 2016 – Present  ·  La Jolla, CA

Staff Software Engineer

  • Principal architect and lead engineer for the GenePattern AI and Notebook ecosystem at UC San Diego.
  • Architected and deployed GenePattern Copilot, a RAG-based agentic AI assistant for GenePattern and bioinformatics research. Built the initial system using LangChain and LangGraph, then migrated to Pydantic AI for its superior structured output handling and data validation.
  • Engineered an MCP server wrapping the GenePattern REST API, enabling Copilot to agentically launch jobs, retrieve results and explore analyses on the user's behalf.
  • Designed and built the GenePattern AI Module Toolkit, a multi-agent pipeline that automates wrapping and deploying bioinformatics tools as GenePattern modules. Uses a hierarchical agent architecture with a supervisor-managed escalation loop across researcher, planner and six artifact-specific agents. Increased GenePattern module production by ~500%.
  • Fine-tuned six Qwen3.5-9B LoRAs — one per artifact type — on LLM-generated, hand-curated synthetic training data at the San Diego Supercomputer Center (SDSC).
  • Led containerization and cloud orchestration via Docker and AWS. Contributed to the development of MSigDB and IGV. Continued architectural leadership of g2nb and GenePattern Notebook throughout.
Python Agentic AI Large Language Models (LLM) Retrieval-Augmented Generation (RAG) Model Context Protocol (MCP) Pydantic AI LangChain Low-Rank Adaptation (LoRA) Vector Databases Embeddings Synthetic Data Generation Prompt Engineering Machine Learning Software Architecture Django Docker AWS Jupyter Linux Cloud Computing JavaScript SQL

Broad Institute of MIT and Harvard

15 yrs 3 mos  ·  Cambridge, MA

Nov 2016 – Present  ·  9 yrs 5 mos

Collaborating Scientist

Joint academic appointment alongside my primary role at UCSD, facilitating cross-institutional open-source development. Lead architect of g2nb, a JupyterLab implementation expanding GenePattern Notebook to integrate Galaxy, Globus and IGV alongside GenePattern.

Jan 2011 – Nov 2016  ·  5 yrs 11 mos

Senior Software Engineer

  • Contributed to a suite of open-source research tools used by over 463k researchers across thousands of academic institutions worldwide.
  • Took full architectural ownership of GenePattern Notebook, a Jupyter-based frontend that renders interactive GenePattern tool interfaces as native notebook widgets. Collaborated directly with the core Jupyter development team, traveling to Berkeley to work with them in person. Built the JupyterHub integration and developed jupyter-wysiwyg, an open-source WYSIWYG markdown editor for Jupyter later adopted by SUSE Linux and distributed via its YaST package manager.
  • Extended the core GenePattern platform — a web-based system exposing hundreds of genomic, bioinformatic and machine learning tools to non-programming researchers. Created its REST API, wrote its Python client library and revised its AJAX frontend. Built its GenomeSpace integration connecting GenePattern to Galaxy, IGV, Genomica and Cytoscape.
Python Java Jupyter Machine Learning Software Architecture Open-Source Software Docker Linux Cloud Computing TypeScript Software Engineering SQL

SkyPilot

Apr 2025 – Present  ·  Remote

Founder & Principal Engineer

  • Solo-founded, built and launched a social media scheduling and analytics SaaS for Bluesky and Threads, selected for the Google Startups Program. Engineered the full stack in Python and Django with Celery and RabbitMQ handling distributed task execution across multiple worker nodes.
  • Built an MCP server wrapping the SkyPilot API, enabling AI agents to schedule posts, manage queues, upload media and retrieve analytics programmatically. Integrated LLM inference for post inspiration and content automation, and ML-based analytics for hashtag performance, engagement trends, optimal post timing and audience growth projections. Profitable every month since launch.
Python Django Celery RabbitMQ Agentic AI AT Protocol Software as a Service (SaaS) Software Architecture Docker JavaScript Software Engineering

Tab Creations

Jul 2013 – Present  ·  Remote

Founder and Creative Director

  • Co-founded an independent tabletop roleplaying game publisher, personally designing and writing four complete games and over 100 published products. Managed the full production pipeline — budgeting, crowdfunding, freelance artists, editors and graphic designers — across 17 successful campaigns raising $135k total.
  • Shadows Over Sol, the company's flagship game, has been licensed for a Portuguese edition and an audio drama produced by GMB Japan. Also designed and built the company website and Foundry VTT integration.
Cross-Functional Leadership Project Management Product Management

Lime Brokerage LLC

May 2007 – Oct 2010  ·  3 yrs 6 mos  ·  Waltham, MA

Software Engineer

Contributed to the trading portal at a high-frequency brokerage platform that processed over 1 billion orders in a single day during the 2008 financial crisis — roughly 9% of total US equity volume that day. Engineered performance optimizations deep in the JSF request lifecycle to maximize throughput with update rates of up to 50 times per second. Took primary ownership of the order book component following a migration to Adobe Air.

JavaScript Java Linux SQL Software Engineering

Education

Brandeis University

2006 – 2007

Master's Degree

Computer Science

Brandeis University

2002 – 2006

Bachelor's Degree

Computer Science

Projects

University of California, San Diego

Sep 2025 – Present

GenePattern AI Module Toolkit

Multi-agent AI pipeline that automates wrapping and deploying bioinformatics CLI tools as production-ready GenePattern modules. Point it at a tool's git repository, documentation and example data, then a hierarchical pipeline of researcher, planner and six artifact-specific agents handles the rest, coordinated by a supervisor agent with structured escalation. Fine-tuned on six Qwen3.5-9B LoRAs trained on synthetic data at the San Diego Supercomputer Center. Increased GenePattern module production by ~500%.

Agentic AI LoRA Pydantic AI Python

University of California, San Diego

Apr 2025 – Present

GenePattern Copilot

Agentic AI assistant for GenePattern and bioinformatics research, combining RAG-based question answering with direct platform control. Built on an MCP server wrapping the GenePattern REST API, enabling the agent to launch jobs, retrieve results and explore analyses on the user's behalf. Migrated from LangChain/LangGraph to Pydantic AI for superior structured output handling and reliability.

RAG MCP Pydantic AI LangChain

Skills

Python Agentic AI Model Context Protocol (MCP) Retrieval-Augmented Generation (RAG) Low-Rank Adaptation (LoRA) Django Pydantic AI LangChain Docker Jupyter Amazon Web Services (AWS) Open-Source Software Celery RabbitMQ SQL Linux Cloud Computing JavaScript TypeScript Java Software as a Service (SaaS) Leadership Product Management Cross-Functional Leadership Project Management Software Engineering AT Protocol

Publications