Sean Sica

In my current capacity as a software engineer, I help lead the software team for MITRE ATT&CK®. Nearly all of my ATT&CK® contributions are visible on GitHub, as most of our tools are free and open source. Most notably, I authored the ATT&CK® Data Model — the first codified expression of the ATT&CK® taxonomy and a TypeScript library for working with ATT&CK® data — and I designed, deployed, and actively maintain our production TAXII 2.1 server ^(source).

I also work on our internal AI Platform team, where my focus has been on mechanistic interpretability (MI) research tooling. I spent significant time contributing to Neuronpedia — an open source platform for AI interpretability research — adding features like Kubernetes support and Azure inference integration (my contributions). Alongside that, I led a research sprint developing in-house tooling to streamline common MI workflows such as SAE training and automated interpretability, which is now published internally for other researchers. That work also led to an exploration of linear probes as a lens for examining how fine-tuning causally shifts learned features (a lightweight alternative to crosscoders). Currently, I'm building an agent-based system to help researchers manage ML experiments and training runs — automating tedious tasks like hyperparameter tuning so they can focus on the science.

Prior to joining MITRE Labs, I was a network engineer in MITRE Corporate, where I was solely responsible for the network infrastructure of our Bedford datacenter. I designed, deployed, and productionized Cisco ACI, establishing the foundation for NetDevOps workflows across the corporate network team — all while completing my undergraduate degree.

Served as the lead integration engineer for a Portsmouth-based MSP, responsible for the full pre-helpdesk client lifecycle: prospecting and securing leads, conducting IT assessments and site surveys, authoring and selling SOWs, and managing end-to-end infrastructure refresh projects. Delivered comprehensive IT documentation and SOP development during client onboarding before transitioning accounts to the helpdesk team.

This was my first real, full-time tech job, working at an MSP. As an intern, I started on the help desk, triaging customer tickets over the phone. I later transitioned to field engineering after acquiring some Cisco certifications. I did a lot of wireless site surveys and infrastructure refresh projects.

Responsible for performing 100BaseT conformance testing against IEEE 802.3 2012 Clause 4 Media Access Control standards utilizing testing techniques and methods outlined in the UNH-IOL Clause 4 Fast Ethernet MAC test suite.

Responsible for performing 100BaseT Ethernet interoperability testing on technologies including switches, end devices, transceivers, etc. Developed sufficient trouble-shooting skills as well as applicable experience in layer 1 and 2 device management.

Responsible for performing Clause 24 (PCS) conformance testing within the Fast Ethernet consortium.

GPA: 3.9.

Formula 1 Safety Car Prediction Engine — DATASCI 210: Capstone (Summer 2025, Berkeley, CA)

• Developed an ML-based early warning system that predicts safety car deployments in Formula 1 races using time series analysis of telemetry data with ROCKET regression, and built the open-source f1-etl Python library for F1 data preprocessing and feature extraction.
• Capstone project selected for Berkeley Summer 2025 Showcase (hover for hyperlink)

Causal Effects of Fine-tuning on LLM Interpretability — DATASCI 266: Natural Language Processing (Fall 2024, Berkeley, CA)

• Conducted original LLM interpretability research analyzing the causal effects of supervised fine-tuning using sparse autoencoders, mechanistic interventions, and representation-level analysis.

GPA: 3.7