Eric Catalano

EECS @ UC Berkeley,
I'm currently a founder at DreamIt, TechOps Director at UC Berkeley IEEE, and creating tasks that help LLM's perform well against terminal benchmarks @ Handshake.

In the past, I did firmware engineering research for Berkeley Formula Racing, and wrote software for all the PCB's on the rocket at STAR.

Work Experience building across software and hardware

Founding Engineer

Full-time

DreamIT · Jan 2026 — Present

Architecting an AI-powered 3D game generation engine built in Rust.
Developing core engine systems including rendering, asset pipelines, and simulation components.
Designing a modular plugin-based architecture to support extensibility and scalable development.
Leading technical execution by defining engineering roadmaps, milestones, and system architecture.
Coordinating development across the team, combining engineering work with project management.

Software Engineer Intern

Internship

DreamIT · Dec 2025 — Jan 2026

Developed internal tooling and prototype features for AI-driven game world generation.
Implemented systems for asset management, rendering, and engine extensibility.
Built demo applications to showcase generative gameplay environments.
Worked with a cross-functional startup team to iterate quickly on new product ideas.

APE Software & Infrastructure Lead

Leadership

UC Berkeley IEEE · Jan 2026 — Present

Leading development of software infrastructure supporting Berkeley's Advanced PCB Engineering initiative.
Building internal tooling and automation pipelines to support PCB design workflows.
Managing repositories, documentation, and development infrastructure for engineering teams.
Collaborating with electrical engineers to streamline board design and testing processes.

AI/ML Research Evaluator

Contract

Handshake · Feb 2026 — Present

Evaluated and refined AI model outputs across multimodal domains including audio, visual, and text.
Assessed generative AI image editing capabilities, providing structured feedback to optimize model precision.
Contributed to iterative model improvement pipelines through systematic quality analysis of AI-generated content.

Avionics & DAQ Engineer

Berkeley, CA

STAR · Jun 2025 — Present

Developing avionics and data acquisition systems for student-built rocket platforms.
Designing sensor boards integrating pressure sensors, thermocouples, and RTDs.
Implementing embedded firmware for microcontrollers handling telemetry and sensor data.
Contributing to rocket flight instrumentation and testing.

Electronics Team

Berkeley, CA

Berkeley Formula Racing · Aug 2025 — Present

Designing custom electronics for Formula SAE electric racecar systems.
Developing a custom 9-axis IMU board with STM32 + ISM330DHCX + LIS2MDL.
Integrating accelerometers, gyroscopes, and magnetometers; writing firmware for CAN communication.
Owned layout decisions around decoupling, grounding, and signal integrity.

Things I've Built from hardware to agents

Witness

Local-first observability for browser agents. Captures DOM diffs, before/after screenshots, the exact action taken, and every LLM call with full prompt, response, and cost, all attached to the step they came from. One line to instrument an agent, then `witness view` opens a viewer on localhost. SQLite plus flat files under ~/.witness/, nothing leaves your machine.

PythonTypeScriptFastAPIOpenTelemetryBrowser Agents

usewitness.dev →

Proof

An honesty layer for coding agents. When an agent claims it ran the tests, built the project, or shipped a working endpoint, Proof actually runs that verification and refuses to ever return a false PASS. Zero-config, pure Python stdlib, with an honesty ledger that tracks every claim and every lie it catches.

PythonAgent VerificationCLIZero-Config

GitHub →

hearth

Security-first NixOS flake for running local LLMs and autonomous agents, sandboxed by default. Agents run as ephemeral, isolated systemd units with no host secrets and no privilege escalation. Every run logs its tokens, cost, latency, and errors to local SQLite, and the entire OS is reproducible from a single flake with atomic, bootloader-level rollback. Ships with a docs site, web command center, and TUI.

NixOSLocal LLMsSandboxingSecurityOllama

GitHub →Docs →

mcp-audit

Scan an MCP server before you trust it. Local-first security auditing for the Model Context Protocol: tool poisoning, hardcoded secrets, command injection, and supply-chain risks, mapped to the OWASP MCP Top 10. Static heuristics plus AST analysis run entirely on your machine, with an optional local-LLM second opinion. Outputs Markdown, SARIF, or HTML.

PythonSecurityMCPOWASPStatic Analysis

GitHub →