EPTHE MECHANICAL COMEDIANPROJECT SHOWCASE
← All hobby projects

Hobby project / Writing & project knowledge

AI-Home-Lab-Writer

An assistant for the work around writing.

A writing project grows into more than a manuscript. Notes, characters, timelines, and decisions need to stay connected as the work changes. The goal is to make that knowledge easier to manage, with AI helping the writer organise it and notice what needs attention.

01 / Intentions

Less administration.
More room to write.

The project aims to simplify managing a writing process by giving AI a supporting role: assistant, organiser, and linker. Finding relevant context, bringing related material together, and pointing out inconsistencies should reduce the effort of maintaining a growing project.

The writer keeps creative authority. The intention is to make decisions easier to understand and revisit, with conflicting information brought forward for review. A model’s suggestion should remain distinguishable from an author’s decision.

Assistant

Find the useful context

Help locate relevant notes and identify open questions so the writer can return to the work with less searching.

Organiser

Keep knowledge maintainable

Help structure project material and highlight information that may need updating as the manuscript develops.

Linker

Make connections visible

Suggest relationships across project knowledge and surface possible conflicts for the writer to assess.

02 / Tools & integrations

A writing workflow on shared inference.

The documented Warden integration uses a Python 3.11 client, asynchronous HTTP through httpx, and a command-line interface. A lease context manager requests a model capability, renews it during long operations, and releases it when the work finishes. Async tests use pytest-asyncio.

Warden manages GPU capacity and runtime placement on the workstation. The Writer client uses the endpoint returned with its lease, so the writing application does not need to manage GPU processes or hard-code workstation ports. Warden can supervise runtimes such as llama.cpp and LM Studio behind this interface.

Multiple inference fallbacks support the broader writing workflow. Their purpose is continuity when a preferred route is unavailable. A different route can change model behaviour and available context, so a fallback needs to preserve the meaning of the task and make its limits clear.

03 / Limits & design priorities

Keep uncertainty visible.

Project knowledge needs human judgement. Apparent contradictions may be deliberate story choices, old drafts, or unresolved decisions. Conflict detection is intended to raise questions for review; it cannot establish which version the author means to keep.

Context and model capacity are finite. An assistant can miss a connection or propose an incorrect one. The design goal is to make suggestions traceable to the project material so they can be checked and maintained.

Unavailable inference is a normal state. The Warden client reports a capacity refusal separately from an operator pause. It does not blindly retry a full workstation or treat a deliberate pause as a transient fault.

The workflow remains the measure of value. The assistant, organiser, and linker roles describe the project’s direction. The Warden client is the documented technical foundation. Success means less effort spent managing knowledge and more confidence in the writer’s own decisions.

Source code & access

Explore the project further.

For access to this hobby project, please send a request on LinkedIn. GitHub access cannot be provided for work projects.

Contact me on LinkedIn