Projects

    Products, experiments, and open-source work. Each entry shows its status.

    Case studies

    Two projects in detail: my role, one decision, the evidence, and the limits.

    Browse the archive

    Puddle

    Status: Experiment
    Checked 9 October 2026

    Does a parent need another feed, or one small next step?

    Audience
    Parents
    My role
    Solo builder.
    Decision
    A playbook of options the parent chooses. Not an open chat that answers for them.
    Evidence
    I checked the public page at askpuddle.com on 9 October 2026.
    Limits
    I did not test onboarding. I have no evidence yet that the guidance helps a family. Puddle makes no medical, developmental, or compliance claims.
    Open question
    Which suggestions do parents try or change?
    Visit askpuddle.com for Puddle (opens in a new tab)
    Screenshot of the SPEAR repository page
    Earlier screenshot.

    SPEAR

    Status: Experiment
    Checked 9 October 2026

    How can an AI application's security behavior be inspected, not just asserted?

    Audience
    Engineers building LLM applications
    My role
    Builder of the open-source middleware.
    Decision
    Put checks at the runtime boundary and publish an evaluation harness. Do not treat provider safety filters as the whole boundary.
    Evidence
    The public repository shows the source tree, README, and evaluation method.
    Limits
    Source access is not a released package or hosted service. The checks cover input screening and deterministic agent controls. They do not cover live-model output leakage or production latency.
    Open question
    How does it hold up with live models, new attacks, and real integrations? The evaluation is limited. It does not guarantee full protection.
    View source for SPEAR (opens in a new tab)

    All projects

    Current work and earlier experiments. I link only to pages I checked. An entry with no check date has no link.

    Status legend

    Experiment:
    A prototype or test. It may not be ready to use.
    Private preview:
    Access is limited.

    32 archive entries

    Screenshot of SPEAR
    Status: Experiment
    Current work · Open Source

    SPEAR

    Secure Prompt Enforcement At Runtime: open-source LLM security middleware with a red-team evaluation harness.

    Checked 9 October 2026

    View source for SPEAR (opens in a new tab)
    Status: Experiment
    Current work · Open Source

    ReviewFlow

    An experimental local review app: a model proposes labels, a person reviews and corrects them, and exports keep each decision, reason, and source.

    Checked 14 October 2026

    View source for ReviewFlow (opens in a new tab)
    Screenshot of Cue
    Status: Experiment
    Earlier experiment · Web App

    Cue

    A personal AI assistant built around Sunny's own daily workflow.

    Screenshot of ACE
    Status: Experiment
    Earlier experiment · Web App

    ACE

    A desktop companion experiment for finding screenshots by what they contain, processed on the Mac.

    Screenshot of Pier
    Status: Experiment
    Earlier experiment · Mobile

    Pier

    An exploration of alerting parents to concerning chat in online games.

    Screenshot of Purrple
    Status: Experiment
    Earlier experiment · Web App

    Purrple

    A habit-tracking experiment with an AI companion cat.

    Screenshot of Voice101
    Status: Experiment
    Earlier experiment · Web App

    Voice101

    An interactive guide to Voice AI terms, concepts and techniques.

    Screenshot of Velocity
    Status: Experiment
    Earlier experiment · Web App

    Velocity

    An experiment in AI-assisted codebase analysis and architecture visualization.

    Screenshot of Personal AI Med Coach
    Status: Experiment
    Earlier experiment · Web App

    Personal AI Med Coach

    A chat prototype for exploring questions about medical topics. Not medical advice.

    Get in touch

    Questions about a project?

    Contact me LinkedIn (opens in a new tab)