Skip to main content

How to Add Human Training To Your Fullview AI Agent

Use this guide to fine-tune your Fullview AI agent for specific support scenarios or to customize how it handles certain interactions.

Dorin avatar
Written by Dorin
Updated over 3 weeks ago

1. What is Human Training?

Human Training lets you teach Fullview AI by example instead of (or in addition to) writing formal Knowledge Center articles. You simply walk the AI through a task exactly the way you want users to experience it—click by click, step by step—so the system can learn from your demonstration, boost its confidence scores, and fine‑tune future answers.

Below is how to manually train with video instructions.

2. Why use Human Training?

Challenge with Knowledge Articles

How Human Training Helps

Articles lag behind product changes and require edits.

Demonstrations can be captured in minutes and keep pace with fast UI iterations.

Written docs may miss edge cases.

Live walk‑throughs naturally include tips, warnings, shortcuts, and context.

Users ask for guidance in many styles.

The AI learns your tone and order of operations directly from your example.

3. How Human Training Works Under the Hood

  1. Capture – You open a training session and perform the workflow for a specific support question or scenario / product workflow.

  2. Encode – Fullview AI automatically converts the session into structured intent–action pairs and aligns them with UI elements.

  3. Weight – A confidence boost is added to the model for matches that resemble your demonstration. More on Confidence Score here.

  4. Evaluate – Each new user request is scored against both existing Knowledge Articles and your Human‑Trained examples.

  5. Adapt – When your interface changes, the AI re‑evaluates element identifiers during inference so training stays relevant without extra work.

4. Step‑by‑Step: Recording a Training Session

  1. Pick the scenario
    Choose a user task where the AI’s current confidence is low or the doc set is thin.

  2. Click on Train and copy the pin code

    1. Pin codes are used to associate specific topics with your trainings.

  3. Launch training mode by long-pressing on the AI widget

  4. Perform the task exactly as you expect the ideal user journey to unfold.

  5. Save the session

Total time: ~2-3 minutes, much faster than drafting a new help article.

5. Best Practices

  • Keep it scoped. One workflow per session; complex branches deserve separate trainings.

  • Retire stale sessions. Archive or replace trainings when a feature is sunset to avoid noise. You can delete trainings from the AI Library any time.

6. Understanding Non‑Determinism (and Why That’s Good)

LLMs don’t replay instructions byte‑for‑byte; they optimize for the user’s intent in real time. You may notice slight deviations, e.g., an extra clarification step or re‑ordering of two harmless clicks. This flexibility allows Fullview AI to:

  • Adjust when your UI labels move or minor wording changes.

  • Personalize explanations to individual user phrasing.

  • Fill gaps with relevant snippets from existing docs if they strengthen the answer.

In short, Human Training sets a high‑fidelity benchmark, while the AI retains freedom to adapt when doing so benefits the customer experience. This helps gain a measurable rise in answer accuracy without sacrificing AI adaptability and avoid manual maintenance.

7. Frequently Asked Questions

Question

Answer

Do I still need Knowledge Center articles?

Yes but they are not the only source of truth.

How many sessions should I record?

Start with your top 20-30 questions, no need to apply manual training to topics that are already answered correctly.

Will the AI override my steps?

No but it may reorder or condense when the user’s intent is clear or slightly different, but core actions remain aligned with your training.

Did this answer your question?