Four builds, and what they do not prove

These are Tahir's independent products and open-source projects. Each writeup separates what the public description or repository documents from what ZUMX cannot verify, because a case study that blends the two is worse than no case study.

Not client work. Read this first

None of the four projects below is a commissioned client engagement. They are Tahir Rasool's independent products and open-source builds, published to show how the work is approached. They do not represent client work, client outcomes or business results, and no such claim is made anywhere on this site.

PKIL

Independent product AI-assisted writing Build 01 of 04

Help someone move from a blank page, rough thought, topic, link or keywords to an editable LinkedIn post draft.

Documented flow

  1. Input a topic, idea, link or keywords
  2. Choose a voice, persona or style
  3. Generate and auto-format a draft
  4. Edit and review it
  5. Scheduling is part of the public feature description

Design and engineering decisions

The draft is the product, not the post

The flow ends at an editable draft the writer still owns. Generation does not publish, and the wording stays the writer's responsibility — which is why editing is part of the documented flow rather than an afterthought.

Voice and persona are an explicit choice

Choosing a voice, persona or style sits next to the topic input instead of hiding in a settings page. The person sees what shaped the draft and can change it before anything is kept.

Drafting and scheduling are separate steps

Scheduling is part of the public feature description, but it comes after review. A draft that still needs a human read is not something that should be able to publish itself.

Product principle

The person stays responsible for their own voice and editing. Generated copy is a draft, not a publishing instruction or a guaranteed result.

RAG ROOT

Independent product Document Q&A and retrieval Build 02 of 04

Ask a question over selected documents and keep the context needed to check the response.

Documented flow

  1. Add documents or PDFs
  2. Ask a question in plain language
  3. Read a generated response beside a source or document view
  4. Cite a page or section so a reader can inspect the material

Design and engineering decisions

The source view sits beside the answer

A generated response is read next to a source or document view rather than behind a link. Keeping the material in view is the difference between a claim you can check and one you can only take on trust.

Citations point at a page or section

A citation that only names a document is decoration. Citing a page or section gives a reader somewhere specific to go, which is what makes inspection practical.

Documents are added deliberately

The person asking chooses which documents or PDFs the system works from. That keeps the scope of what can be answered visible and deliberate.

Product principle

Source inspection is part of the experience. A citation helps a human check a response but does not guarantee completeness or correctness.

MotionForge

Open-source project Programmatic video tooling Build 03 of 04

Create a reusable authoring-to-render workflow for code-driven video and motion, instead of rebuilding every scene as isolated animation glue.

Repository-documented scope

  • React / Next.js and TypeScript workflow
  • WebGL / Three.js effects
  • Reusable motion and effect components
  • Spring physics and 3D transforms
  • Visual timeline editing
  • Frame-accurate rendering
  • WebCodecs video export
  • An HTML-native HyperFrame path
  • CLI scaffolding

Design and engineering decisions

A timeline you can see, not only a config file

Visual timeline editing is part of the documented scope, so the sequence being rendered is inspectable in the tool rather than only in code. That is what makes a frame range reviewable.

Determinism before throughput

The engineering focus is deterministic output at a known frame and time. Browser, font, codec, resolution and long-composition behaviour are treated as validation work — the parts that decide whether a render matches the previous run.

Two output paths, one authoring model

The documented scope pairs a WebCodecs video export with an HTML-native HyperFrame path. Same scene description, different destination, which keeps the authoring work independent of the export path.

Engineering focus

Deterministic output at a known frame and time. Browser, font, codec, resolution and long-composition testing are treated as validation work before any production throughput or fidelity is promised.

ClipLite

Open-source project Desktop clipboard utility Build 04 of 04

Retrieve an older copied link, snippet or text after a newer clipboard item replaces it.

Documented flow

  1. Cross-platform clipboard history
  2. Global shortcut Ctrl+Shift+V
  3. Searchable history
  4. Pin items
  5. Keyboard navigation and selection
  6. Paste the selected item
  7. Local database

Repository-documented stack

Tauri v2React 18TypeScriptRustSQLite

Repository-documented reliability work

  • Use the platform app-data directory for the database path
  • Parameterize search queries
  • Handle shortcut collisions without crashing
  • Retrieve older clips by ID
  • Clean up async listeners
  • Accessibility and reduced-motion handling described in project notes

Design and engineering decisions

Retrieval is the product, not the capture

The clipboard history is easy to write and hard to make useful. Getting an older clip back by ID, searching parameterised queries, pinning and keyboard selection are what turn a log into a tool.

The database lives where the platform expects it

Using the platform app-data directory for the database path avoids writing into a location the OS or user may later treat as disposable.

A global shortcut must not be able to crash the app

Shortcut collisions are handled rather than assumed away, and async listeners are cleaned up — the two places a desktop utility of this kind most easily leaves a stale handler behind.

Direct links to live product pages and individual repositories are intentionally omitted. The verified public profile is Tahir's GitHub and LinkedIn. Guessed URLs would be worse than none.

Individual repository URLs are not confirmed in the canonical profile either. Both profiles are linked here: GitHub and LinkedIn.

Want the same care applied to your problem?

The four builds above show the approach, not a client roster. Bring your own problem and it gets the same written scope and the same reviewable process.