About int8ai

AI engineering, from models to systems

Author

int8ai is written and maintained by Palak Kala, a software engineer working on AI systems and infrastructure. The goal is to build a visual field guide that helps engineers grow toward Staff and Principal AI engineering capability — lessons where a reader can explain a mechanism, inspect a real implementation, and make a better engineering decision.

No institutional affiliation, certification, or formal credential is claimed.The lessons reflect personal study, implementation, and engineering judgment. Readers should verify claims against primary sources and their own measurements before applying them in production.

Publication purpose

Free, original lessons on model internals, inference infrastructure, and AI systems engineering — no paywalled content, no affiliate links, no sponsored material.
Evidence-based explanations tied to primary sources (papers, official documentation, reproducible measurements) rather than vendor claims or secondary summaries.
Staff/Principal competency focus: each lesson targets a concrete engineering outcome — trace tensors, estimate memory/latency, choose a design under explicit constraints, write an architecture decision record.
Owned audience, no platform dependency: the site is a static Astro/Starlight build on a custom domain. Distribution starts with search and direct visits; a newsletter may be added later only after demonstrating reader value.

Editorial and source-review process

Every published lesson goes through the following stages before release:

Research brief — a source-mapped brief is produced for each topic, citing original papers and official documentation with exact section references.
Lesson draft — the brief is adapted into a lesson with worked examples, original diagrams, and interactive components where they clarify a mechanism.
Independent technical review — a separate review pass checks every causal and performance claim against its cited source, verifies arithmetic, inspects diagrams for accuracy, and validates interactive behavior. Reviewers do not approve their own work.
Editorial pass — clarity, accessibility (WCAG 2.2 AA target), and consistent style are verified.
Reader test — where feasible, a small group of target readers completes the lesson and reports confusion points before publication.

The two inference primers have completed an initial focused review. Additional Track 1 and Track 2 lessons have draft pages and some have received separate technical reviews or corrections. The full tracks are still undergoing a lesson-by-lesson release review. Pages available here should be read as drafts until that review is complete.

Corrections and contact

No verified contact channel is currently published. A corrections workflow, public issue tracker, or email address will be added before broad promotion, as required by the launch readiness gate (plans/launch-readiness.md).

Until a dedicated channel exists, readers who spot an error can:

Open an issue in the site repository (if public) with the lesson URL, the specific claim, and a suggested correction with source.
Reference the PILOT_LESSON_REVIEW.md format: severity, location, finding, and proposed change.

All substantive corrections will be acknowledged in a public corrections log once the channel is established. Typos and formatting fixes may be applied silently.

Licensing and reuse

Lesson text and diagrams: CC BY-NC-SA 4.0 — attribution required, non-commercial use only, share alike.

Code examples in lessons and labs: MIT License — use freely in your own projects.

The site build, configuration, and custom components: not licensed for reuse; they are the implementation of this publication.