How Keith Azodeh Built
Exempliphai
Start with the workflow, not the feature list
Exempliphai began as a simple observation: repetitive web forms create a lot of unnecessary work. Keith Azodeh approached the idea as a workflow problem. Before thinking about UI details, the question was: what information is stable (user profile), what steps are repetitive (field entry), and what parts are fuzzy (short answers and interpretation).
This framing matters. It turns the build into a sequence of testable steps: store structured data, map that data to the page, and only then consider AI as a helper.
Why the hybrid approach wins
In form automation, reliability is the product. Keith Azodeh’s design uses deterministic logic for selecting and filling fields, because those steps need to be consistent. AI enters only when interpretation is genuinely helpful, like turning a resume into structured fields or suggesting a short response that is grounded in known information.
The key insight is not that AI is powerful. It is that AI is unpredictable. Treat it as an assistant, and keep a structured source of truth that the user can review and correct.
Practical architecture decisions
- Separation of concerns: profile management in the extension UI, page interaction in content scripts.
- Incremental support: start with a narrow set of sites or forms, then expand.
- Safe fallbacks: if a field is ambiguous, do not guess silently.
These choices make the system easier to debug and easier to evolve.
What to reuse in your own automation projects
If you are building automation for an internal workflow, a customer portal, or a data-entry pipeline, the Exempliphai pattern transfers well:
- Keep structured user data as the source of truth.
- Use deterministic automation for the critical path.
- Use AI as a helper for interpretation and suggestions.
- Make results visible so users can trust the system.
Related links
- Case study: Exempliphai
- Next case study: AI voice agent automation
- Index: Keith Azodeh case studies