bryce cole

hey i'm Bryce. I'm currently a Software Engineer at Yahoo, building AI features for a cross-platform browser.

Previously, I worked at BlackRock on AI-assisted compliance and internal research tools. I'm interested in building dependable coding agents — the kind of work I'd be excited to do with the Codex team at OpenAI.

Updated by Codex· July 10, 2026

An agent reviews my project conversations and work across my devices, then periodically updates this site with relevant experience and projects.

Featured work

End-to-end systems where model behavior, tools, data, and product surfaces have to work together.

Research harness / paper trading

Kalshi BTC Perps Monitor

local build

I built this to test whether a short-term BTC strategy was worth trusting before it ever touched real money.

It pulls live Kalshi data, signs requests, simulates long / short / flat trades, and records every decision. A strategy only graduates from paper mode if it survives fees, slippage, and multi-week holdouts; live ordering stays disabled by default.

PythonKalshi RESTwalk-forward evalspaper mode

Native iOS / personal research loop

Brain Feed

source ↗

I built a personal feed because generic feeds kept mixing useful research with noise and forgot what I had already seen.

It collects sources into an offline SwiftData queue, remembers read / save / like / skip decisions, refreshes in the background, and notifies me about high-severity posts. Sources open inside the app so the research trail stays intact.

SwiftUISwiftDataCloudflare Workerbackground refresh

Device systems / streaming reliability

iPhone SE SecurityCam

source ↗

I repurposed an old first-generation iPhone SE into a local security camera instead of buying another camera.

The phone captures H.264, serves an RTSP stream, advertises itself over Bonjour, survives a locked screen, and exposes health checks. Scrypted rebroadcasts it into HomeKit, with a frame watchdog and thermal guardrails for the parts that fail in real use.

Objective-C++RTSPBonjourScrypted / HomeKit

Agent operations / recovery

Hermes Automation Layer

source ↗

I built the layer around my scheduled agents because a green cron job does not mean the useful work actually happened.

Hermes shows gateway pulses, model routes, cron state, browser-tool boundaries, Pi-side checks, and durable run notes. Jobs report proof or a specific failure class, then recover through an explicit path instead of silently staying stuck.

PythonlaunchdMCPoperational evals

Data ingest / signal verification

Market News Truthness

local build

I built this because a headline is not a trade thesis until the underlying price and volume action agree with it.

It collects six-plus news providers, hashes and archives each headline in JSONL, deduplicates repeats, and sends selected items through hypothesis generation plus price / volume checks. Collection and analysis stay separate so every alert has a traceable source.

Pythonmulti-source ingestJSONL archivesignal verification

Blog

Recent writing.

Earlier product experiments

A few smaller products and interface studies from the same build-and-learn loop.

Local experiments

Small, opinionated systems built on the machine or against real devices, where the interesting part is the feedback loop.

K36 Terminal Launcher

I built a terminal-style game selector because the stock K36 launcher made a tiny handheld feel slower and noisier than it needed to be.

It scans the real ROM mount, launches through ArkOS's existing wrappers, and returns cleanly on Select+Start. The staged boot text, held-input repeat, and repaint logic came from testing the thing on the actual device.

Safari Sync Bridge

I built this because I browse in Chromium but still want Safari and iCloud to be the source of truth for my Apple devices.

The extension and native host reconcile bookmarks, Reading List, tabs, tab groups, and history locally. Copied-file tests, reversible backups, generation-aware History.db writes, and a Helium/Safari canary keep replay safe.

Kitty Mode

I built this for a very specific roommate problem: my cat likes to nap on the Mac when it is warm, and his paws can click or type into whatever is open.

Kitty Mode locks out keyboard and media-key events while he warms himself on the laptop, with a menu-bar toggle and triple-tap Right Option escape hatch. Accessibility checks and suppression counts make it clear when the cat-proofing is actually active.

Study Guide Workspace

I built this to turn a vague study question into a usable plan instead of a pile of tabs and half-finished notes.

A topic prompt becomes an overview, highlighted concepts, deeper subtopics, and a quiz. SwiftUI state, keyword extraction, progress tracking, and UI tests keep the loop useful on-device.

Document Censor

I built this for the uncomfortable moment before uploading a tax form, application, or other PDF to an AI tool or sending it to someone else.

It redacts visible text and form fields, strips metadata, bookmarks, scripts, and embedded files, then scans every PDF layer for leftover PII before the cleaned copy leaves the machine.

Working systems

Visual snapshots of the systems I have running, what they watch, and what they are trying to automate.

TradingAgents Command Center

A daily market-research workspace that runs scheduled feeds, turns them into Discord-ready analysis, and saves the report locally.

I built it to make the watchlist repeatable and to keep buy, sell, trim, and hold decisions from disappearing into chat.

Brain Feed Curator

The selection pass behind Brain: it turns market notes, source checks, and my reading behavior into a short queue.

I built it to stop the feed from becoming another firehose, using recency, deduplication, and a judge pass before anything is queued.

Hermes Automation Layer

A status surface for the local agents that run through cron, a menu companion, and Raspberry Pi experiments.

I built it so I can tell whether an agent is healthy, waiting on a tool, or needs recovery without opening five terminals.

Hermes Siri Chat

A phone-style SwiftUI chat for asking the local Hermes agent what changed and telling it what to run next.

I built the Mac bridge and LAN fallbacks so a useful agent session survives across devices instead of living only in a terminal window.

iOS Tweak and Deploy Loop

A personal build-and-install loop for jailbreak tweaks and sideloaded iOS experiments.

I built it to shorten the path from a UI idea to a real device: package, sign, install, respring, launch, and check for crashes.

iPhone SE Security Camera

The HomeKit-facing half of the old-iPhone camera experiment.

I built the Scrypted bridge so an improvised RTSP stream behaves like a normal door camera, with direct-stream checks, Bonjour-safe routing, clip health, and non-destructive recovery.

Native Terminal Experiments

A study in making remote shell sessions feel like a real iOS app instead of a web terminal in a box.

I used it to work through tabs, SSH sessions, hardware-key mapping, scrolling, and rendering glitches that only show up during actual device use.

Personal Notes App

A local SwiftUI workspace for the notes I actually use: market research, device checklists, and things to read later.

I built it around notebooks, reminders, search, and local storage so capturing a thought does not require opening a full productivity system.

Institutional Flow Agent

A screener watcher that turns unusual fund activity into a short list of names worth investigating.

I built it to rank activity by conviction and context rather than treating the biggest raw dollar amount as the most important signal.

Market News Scanner

An hourly news collector that turns a stream of headlines into tagged, searchable market events.

I built it to extract entities, remove duplicates, score likely impact, and preserve the source before analysis changes the wording.

Polymarket Monitoring Agent

A watcher for prediction-market odds and repeat-wallet behavior.

I built it to distinguish a real shift in informed activity from a one-off price move, then connect unusual trades back to outside events.

Archive and Research Agent

A local archive for market data, scraped context, and the signals my other agents produced.

I built it so a new report can retrieve what happened last week or last quarter instead of starting from a blank prompt every time.

iOS Deployment Agent

A small build queue for watching personal iOS archives, signing, installs, and launch smoke tests.

I built it to remove the repetitive device work from app iteration and make a failed install distinguishable from a failed app launch.

Housing Search Agent

A practical search assistant for listings, application requirements, and the deadlines that come with moving.

I built it to keep rent, tours, documents, and follow-ups in one place instead of losing the useful details across tabs and messages.

Experience

AI product engineer focused on browser intelligence, cross-platform apps, automation, large-scale data pipelines, and LLM-backed product workflows.

Yahoo

Software Engineer, Browser AI

Apr 2026 to Present

Building AI features for a cross-platform Chromium-based browser spanning desktop, iOS, and Android, with ownership across page-content understanding and personalized daily feed creation.

Turn browser context into useful product surfaces: extracting signals from pages, shaping model-backed summaries and recommendations, and making AI feel native inside everyday browsing flows.

LabelBox & Anthropic

AI Code Evaluator

Dec 2025 to Apr 2026

Review and improve multi-file code changes generated by Claude across real repositories, including feature work and bug fixes in established open-source codebases.

Contribute directly to coding-model quality through RLHF-style evaluation, fine-tuning feedback, and synthetic data generation across Python and Rust tasks.

Simplify

Backend Software Engineer

Mar 2025 to Oct 2025 · Remote

Maintained and improved an AI parsing pipeline processing 50 to 60 thousand job postings daily, refining prompt logic and type validation to reduce hallucination-driven data errors.

Built SEO-focused Next.js job pages and fixed SQLAlchemy n+1 issues, cutting response times from roughly 1.9s to 100 to 300ms while improving search discoverability.

BlackRock

Software Engineer Intern

Jun 2025 to Aug 2025 · New York

Built an automated KYC compliance system that searched SEC, DOL, FFIEC, NAIC, OpenCorporates, and other sources through multithreaded API, scraping, and agent workflows.

Reduced manual compliance review from about 3 hours to 5 minutes with document retrieval, confidence scoring, and cited AI-assisted verification for internal teams.

BlackRock

Software Engineer Intern

Jun 2024 to Aug 2024 · New York

Built Angular components for an internal Aladdin research chatbot and used RxJS-driven streaming to preserve screen and widget context in responses.

Reached 99% intent recognition accuracy for finance-specific queries with a custom NLP classification pipeline and multi-stage prompt routing.

Apple

Product & Technical Specialist

May 2022 to May 2024 · New York

Provided expert macOS and iOS support, resolved complex customer issues, and maintained a 99% satisfaction score while improving store throughput.