Software Engineering Intern · Wrodium
Jan 2026 – Mar 2026- Built content refresh tooling that parses external URLs, detects stale mirrored content, and generates reviewable change proposals
PythonTypeScriptPostgreSQLScrapingRAG
Matt Okner
EECS @ UC Berkeley · ML, agents, and systems.
Berkeley, CA
I build machine learning systems, real-time AI agents, and applied software that turns research ideas into working infrastructure. Lately I have been thinking about stablecoins, probability, and on-chain incentives.
Experience
Member · Blockchain at Berkeley
Aug 2025 – Present- Exploring agent-to-agent payments, on-chain infrastructure, and verifiable compute systems
Smart ContractsConsultingResearchOn-chain Systems
Research Assistant · San José State University
Jun 2023 – Aug 2023- Trained a malware classification MLP achieving 91% accuracy on 15k PE-file samples
Pythonscikit-learnMalware AnalysisMLP
Projects
Featured builds, not the whole archive.
Building · 2026
Warden
Policy layer for agents before tools run.
Details
Runtime guardrail for OpenAI Agents SDK tools. It checks tool names, schemas, and arguments, then allows, blocks, or asks for approval before side effects happen.
- Tool boundary: Wraps an agent's tool array without changing the agent loop
- Policy engine: Allows reads, gates writes and external sends, blocks secrets and payment-like calls
- Approvals: Pauses risky actions for Telegram or custom review
- Audit trail: Writes JSONL decisions with redacted arguments
- Setup: Scaffold a guarded agent or wrap existing tools with guardTools()
Building · 2026
miniMoE
Training a GPT-2-level LLM cheaply with sparse MoE.
Details
A compact MoE language model aimed at GPT-2-level capability on a cheaper sparse-training budget.
- Model: Transformer blocks with top-k routed expert feed-forward layers
- Data: Shakespeare text loader for tokenizer experiments
- Goal: Use MoE routing to reduce dense training cost
ETH · 2026
Procrastinot
USDC commitment stakes with oracle-judged evidence.
Details
Commitment app where users stake USDC, submit proof before a deadline, and receive a pass, retry, or forfeit outcome from an oracle.
- Contracts: Foundry contracts for commitments, attempts, verdicts, forfeits
- App: Next.js wallet flow for staking and evidence upload
- Oracle: Evaluates evidence and writes pass/fail outcomes on-chain
- Indexer: Syncs contract events into Supabase for fast reads
- Settlement: USDC escrow with retry and forfeit paths
Built · 2025
Sage.ai
Real-time phone agent for booking, intake, and triage.
Details
Voice AI for inbound and outbound calls where response time and reliable handoff matter.
- Latency: Targets sub-500ms speech-to-speech turns
- Transport: Twilio Media Streams over WebSockets
- Speech: Deepgram STT with low-latency synthesis
- Agent loop: Routes intents and tool calls through Anthropic
- Deploy: Runs on Railway with graceful session cleanup
Show 4 more projects
Evaluations · 2026
CharCounting
Eval for when LLMs should stop guessing and use tools.
Details
Compared GPT-4o character counting with direct reasoning versus a Python interpreter. Tool use removed the measured counting errors.
- Baseline: 77.8% accuracy without tools on repeated or long strings
- Tool run: 100% accuracy with a Python interpreter
- Finding: Token boundaries cause avoidable counting mistakes
- Harness: Reproducible eval prompts and result checks
Built · 2026
Zetamac, but Better
SwiftUI mental-math trainer with an AI coach.
Details
Native iOS arithmetic trainer with zetamac-style drills, local history, and an AI coach that turns recent solve times into practice plans.
- Platform: SwiftUI app for iOS 17+
- Profiles: Local users and score history in UserDefaults
- Coach: Anthropic-powered feedback from recent solve times
- Drills: Zetamac-style presets by operation and difficulty
- Stats: Tracks streaks, speed, and accuracy over time
Presented · 2023
PE Malware MLP
91% accuracy malware family classifier on 15k PE-file samples.
Details
Static Windows PE-file classifier trained on a curated VirusTotal slice and presented at the SJSU Research Symposium 2023.
- Model: Three-layer MLP over static PE-file features
- Data: 15,200 VirusTotal samples across 8 families
- Features: Imports, section metadata, headers, entropy
- Result: 91.3% test accuracy with per-family precision/recall
Deployed · 2022
Smart Irrigation System
Raspberry Pi + Flask automated garden controller.
Details
Raspberry Pi controller that reads garden sensors, drives solenoid valves, and logs moisture data through a small Flask API.
- Sensors: Capacitive soil probes plus temperature and humidity
- Control: Schedule and threshold watering modes
- Valves: Solenoid control with manual override
- Logs: SQLite moisture and run history
- Result: About 30% water reduction in a 200 sq ft test garden
Education
UC Berkeley
Aug 2025 – May 2027B.S. Electrical Engineering & Computer Science
GPA 4.0Honors Scholar
CourseworkView courses
- CS 61AA
- CS 61BA+
- CS 61CA
- DATA 8A
- CS 70A+
- EECS 16AA
- EECS 126A
- CS 189A
West Valley College
Aug 2023 – May 2025A.S. Mathematics / A.S. Data Science / A.S. Physics
GPA 4.0Summa Cum Laude
CourseworkView courses
- MATH 3AA
- MATH 3BA
- MATH 4AA
- MATH 4CA
- MATH 4BA
- PHYS 4AA
- PHYS 4BA
- PHYS 4CA
- CHEM 1AA
Contact
Let's build something.
Open to internships, research collaborations, and serious side projects. Email is the fastest way to reach me.