Corbin Floyd

Hi, my name is

Corbin Floyd

AI Researcher, Evaluator, & Builder

See my work Work history
Corbin Floyd

Why me?

I Love
solving complex problems!

I evaluate AI systems, build reusable agent skills, and write technical documentation. I came to this work through financial support and business technology consulting.

Read my story

Pensacola, FL

AI evaluation Rubric grading Model output review Agent workflows Multi-agent systems Evaluation harnesses A/B testing Graph memory Prompt evaluation Technical documentation MCP servers Agent skills Fraud and identity review Research synthesis

What I Stand For

I think advanced AI can improve people’s lives. Its rollout needs safeguards, reliable infrastructure, public understanding, and institutions prepared to handle it.

The benefits need to reach ordinary people, too. If a small group controls the models, infrastructure, and decisions, those benefits will slowly diminish.

Pensacola, Florida AI Fellow Researcher Builder Consultant Innovator Business technology AI evaluation Financial support

Work record

Work History

I now focus on AI evaluation and applied AI projects.

  1. I Heart Bento

    Restaurant Manager

    I managed day-to-day operations and a nine-person team, including training, scheduling, inventory, and opening and closing.

  2. Navy Federal Credit Union

    Financial Advisor (MSR II)

    I helped members with account access, fraud concerns, and identity checks. I documented cases and escalated issues that needed further review.

  3. Vicar Group

    Technology and Innovations Advisor

    I researched AI, cybersecurity, fraud, fintech, and business technology, then prepared source-based guidance for business discussions.

  4. Handshake AI Fellowship

    AI Fellow - Trainer & Reviewer

    I evaluate AI outputs and review other fellows’ submissions.

Projects

  1. 01Guidance for coding work, checks, and handoffs

    Agentelligence

    Plugin development · 2026 · Private developer preview

    Agentelligence is a private Codex plugin I’m developing for longer coding jobs. It adds instructions for checking project context, dividing work, and verifying the result.

    Agentelligence task flow: a requested outcome and target project enter the parent skill, which selects useful guidance. Codex performs the work using the project’s instructions and tools. Optional workers return files and findings to the parent. The parent inspects the result, continues work if needed, and reports what is observed or unresolved. An optional handoff transfers unfinished work to another task.
    The entry skill selects only the guidance a task needs. Codex performs the work; delegation and handoffs are optional.

    Stack: Codex, Agent Skills, evaluation fixtures

  2. 02Multi-agent stock research and separate trading controls

    TradingAgents

    Framework extension and research · 2026 · Private research project

    TradingAgents is my extension of TauricResearch’s multi-agent research framework. I added overnight research, stored evidence, decision tracking, and a separate layer of trading controls.

    TradingAgents extends TauricResearch TradingAgents. Four analyst reports cover technical signals, sentiment, news and company fundamentals. Bull and bear researchers debate the reports, a research manager develops a plan, and a trader proposes an action. Aggressive, conservative and neutral risk reviewers debate the proposal before the portfolio manager records a decision. Stored research context and crowd-scenario records are supporting inputs, not mandatory decision stages. MiroFish reports and local scenario records provide advisory hypotheses, not trade instructions. Saved decisions can be compared with later outcomes for reflection on a future run of the same symbol. Broker execution is separate from the AI graph: it requires authorization and code-based policy and risk checks before a paper or gated live broker action.
    Analyst reports pass through research debate and risk review. Stored context and crowd scenarios support the research; broker execution has separate authorization and code-based checks. Built on TauricResearch TradingAgents.

    Stack: TradingAgents framework, research memory, paper and broker workflows

  3. 03Finding and comparing tools for AI coding tasks

    Skill Finder

    Solo build, public release · 2026 · v1.3.0, MIT licensed

    Skill Finder is a plugin for Codex and Claude Code that finds and compares tools for a task. Describe what you need, such as browser control or support for a file format, and it searches for existing skills, plugins, connectors, and packages.

    Skill Finder: define the task, find candidates, verify sources, compare options, and recommend a next step. Weak sources send the search back to another route. Insufficient evidence produces a provisional shortlist or a missing-capability specification.
    Skill Finder checks search results against source evidence, compares the options, and keeps uncertain findings visible.

    Stack: Codex, Claude Code, Agent Skills

    View the v1.3.0 release

  4. 04Comparing drafts to improve AI writing instructions

    Voice Tuning Lab

    Personal lab · 2026 · Private

    Voice Tuning Lab is a local web app I built to improve the writing instructions I give AI assistants. It presents several drafts of the same piece and records which versions I prefer.

    Voice Tuning Lab: compare Candidate A, B, and C; mark choices, edits, and phrases; save the decision; check patterns across tasks in the question ledger; and propose instruction edits for review. Unsettled or conflicting feedback returns to more comparisons. The lab does not automatically change the writing skill.
    Saved writing preferences feed a question ledger and proposed instruction changes. Unsettled questions return for more comparisons.

    Stack: Local web app, scenario runs, grading rubric

  5. 05Smaller tools and experiments

    Other Work

    Public repositories · Ongoing

    My GitHub includes smaller tools, experiments, and releases alongside the projects shown here. The repositories are the best place to inspect their code and documentation.

    Stack: Mixed

    View my GitHub

    View MoneyMap

    View Fable 5 Prompt Optimizer

    View Tower Upgrade Advisor

Files

Certificates

Exhibit 02 · Claude 101 Anthropic course completion
Email LinkedIn GitHub Handshake

Contact

Get in touch with me

Email is the fastest way to reach me.