Chai With JaiGPU Study Hall
Menu
Courses / Curriculum / A1
FREE TIER · OPEN ACCESS ⏱ 15 mins·Orientation

First Action and Source Grounding

How do you make sure an AI assistant uses your real notes instead of guessing?

Ground your draft in real facts. Save an exact source excerpt, record its file path and line numbers, and pull out confirmed facts before you write your prompt. Keep authored practice fixtures separate from your personal task card.

Before you start

Keep open your personal return_card.txt from lesson A0 (or your saved browser entry at /courses/workspace), and ensure you have a terminal or text editor ready for local files.

Grounding Prevents Guesswork

Language models predict words based on common patterns across the internet. When you ask an assistant to draft a resume, write a proposal, or summarize a meeting without giving it the raw text, it fills the missing gaps with plausible assumptions. It might invent job titles, make up software tools you never used, or state deadlines that do not exist.

Grounding means giving the model the exact source text first and requiring every statement to come directly from those lines.

The Three Parts of Source Grounding

To keep your draft grounded, record three details before you run your prompt:

  1. File Location: The exact path where the source file lives on your computer.
  2. Line Locator: The specific line numbers, section title, or timestamp.
  3. Confirmed Facts: The exact dates, names, or numbers that the draft must match.

Separating Practice Fixtures from Personal Work

When learning these steps, keep supplied practice exercises in a separate folder (such as course-practice/) so you do not mix fictional sample data into your real projects. Your personal Return Card from lesson A0 belongs to your own task; never copy supplied case data into your personal Return Card.

Worked Example: Extracting Facts from Notes

An operations manager prepares notes for an onboarding document. Instead of pasting an entire inbox, the manager creates a clean source file.

First, create a folder named course-practice/sources and save this file as course-practice/sources/account_notes.txt:

Line 1: Account setup for freelance designers.
Line 2: Design files live in Figma under the Marketing workspace.
Line 3: Invite new hires as Viewers first. Team lead upgrades to Editor on day two.
Line 4: Track hours in Harvest under project code MKT-402.
Line 5: Submit time sheets every Friday by 5:00 PM EST.

Next, the manager records the source details on a practice card (course-practice/practice_card.txt):

  • File Location: course-practice/sources/account_notes.txt
  • Line Locator: Lines 2 through 5
  • Confirmed Facts:
  1. Figma workspace name is Marketing.
  2. New hires begin as Viewers; upgrade to Editor happens on day two.
  3. Harvest project code is MKT-402; timesheets are due Friday by 5:00 PM EST.

When the manager prompts the assistant, the prompt includes lines 2 through 5 and tells the assistant not to mention any software tools outside those lines. This gives you specific facts to check in the draft. The assistant can still make mistakes, so compare its claims with the source before using them.

Try it

Complete the two separate parts below. First work through the supplied practice fixture, then ground your own personal task.

Part 1: Supplied Practice Fixture

  1. In your terminal or file manager, create a folder named course-practice/sources:

mkdir -p course-practice/sources

  1. Create course-practice/sources/project_facts.txt and paste in these exact lines:
Line 1: Client project: Oakridge Community Portal launch.
Line 2: Staging review deadline is Thursday, November 14 at 3:00 PM.
Line 3: Primary contact is Marcus Vance (marcus@oakridge.example).
Line 4: Testing server URL: staging.oakridge.example:8080.
Line 5: Sign-off requires approval from both Marcus Vance and Elena Rostova.
  1. Create a practice record in course-practice/practice_source_record.txt with:
  • File Path: course-practice/sources/project_facts.txt
  • Line Numbers: Lines 1 through 5
  • Three Confirmed Facts: [Extract 3 verifiable facts directly from those lines]

Part 2: Your Personal Task Grounding

  1. Now open your personal project folder from lesson A0 (or open the Source stage in the workspace). In your personal project folder, create a sources/ subfolder (for example, mkdir -p my-project/sources). Keep your personal task from lesson A0 intact; do not erase it.
  2. Save 5 to 10 lines of real source text from your own actual project into my-project/sources/my_notes.txt.
  3. If working in the web workspace, open Source and scroll below the map to Source draft and notes. Record your personal source location, line numbers, and three confirmed facts together in that field. Click Save a version, return to this lesson, then reopen your Source notes to check they remain. If working locally, record your personal source location, line numbers, and three confirmed facts in my-project/return_card.txt. Do not copy Oakridge practice facts into your personal task or source notes.

Try a different case

Inspect your personal source file (my-project/sources/my_notes.txt). Verify that each of the three confirmed facts in your personal Return Card can be highlighted on a specific line of that file. If any fact contains an assumed detail or outside number, edit your card to reflect only what is explicitly written.

Save your work

Two separate verified items: (1) course-practice/sources/project_facts.txt and course-practice/practice_source_record.txt for the practice run, and (2) your updated personal return_card.txt (or saved /courses/workspace entry) citing your personal source file path and confirmed facts.

Next

If your personal Return Card contains a verified local path, line locator, and three facts drawn directly from your own source file, proceed to Crash Course lesson C1 to draft your deliverable. If your personal card contains Oakridge practice details or unverified facts, fix them before moving forward.

Apply in Your work → All lessons