Enoch AI / Lab

Under the hood.

The technical side: personal projects, how I use AI coding tools, and what I learn by building and using the results.

Personal project

DocLifts — a live workout app with AI assistance

A live multi-user Alpha built around real training: photo-assisted machine identification, machine-specific history, editable programs and four workout views. Read-only MCP access is live, with account consent and revocation. Muse has read app workouts and imported notebooks; bulk history reads keep the request count down.

The latest release, 0.20.0, adds JSON and spreadsheet exports, plus an experimental Strong-style CSV. Account deletion and saved-photo removal are also live. Photo files and restoring exports remain future work, and a real Hevy import still needs verification. Alpha access is operator-managed; public signup is closed.

Read the process and see the app →

Agentic security updates

October 30, 2026 · Field notes

An alert on X became a coordinated security review across the Linux host, application, containers, database and Cloudflare Tunnel. The article follows the tested fixes, the brief deployment interruption and the findings still open.

Read the article → All writing

From security review to shipped fixes

October 24, 2026 · DocLifts

A second security review surfaced eight leads. We reproduced four behaviors and shipped fixes for login throttling, OAuth refresh, backup permissions and maintenance credential handling. A separate Node/Brotli security update patched both the app and cleanup worker. The write-up follows the process: challenge the findings, reproduce the behavior, review the changes and verify production, while keeping unresolved follow-ups visible.

Read the security review and fixes →

Tools

  • Stack →

    What I'm building with right now: hosting, app stacks, models, agent runtime, workflow infrastructure.

How I work

A recent personal build went from planning to production using a coordinated multi-LLM workflow: separate tools for design review, implementation, code review, and test authoring, all kept aligned through version-controlled docs instead of shared memory. The human stayed in the loop for decisions and real-world verification. Claims were checked, not trusted.

That discipline is the point. The interesting part is not which model you use; it is the operating system around the tools.

Case study: DocLifts — Document your Lifts →

Currently building

I’m building personal tools and experimenting with coding agents and automation. DocLifts is now live with consent-based agent access and downloadable training data. The next priorities are a smoother first workout and clearer separation of test entries from real training. Phone and gym acceptance will guide the onboarding work; replacement-server recovery and capacity checks remain ahead of wider Alpha access.

I’m also exploring private, access-controlled tools in the cloud for a particular person or team. The focus is a useful workflow with clear permissions and a way to check the result.

Elsewhere

Talk through a technical task

Have a workflow or integration you want to work through? Bring the constraints and what you've tried.

© 2026 Enoch AI LLC · Delaware · Privacy