Your first AI project: a research brief you can check

A good first AI project has a narrow scope and an answer you can inspect. Comparing two supplied venue descriptions is a better learning exercise than asking for an entire business strategy with no evidence to judge it against.

This exercise uses fictional information. You can review the supplied draft without an AI account. If you have an approved tool, you can also generate your own comparison and record the difference between operating the tool and checking someone else's output.

The task

You are planning a 20-person workshop. Produce a short comparison using only the supplied capacity, access information and hire terms. Identify unanswered questions and explain whether either venue can be confirmed yet.

Success means keeping the facts and their limits intact. It does not mean choosing a winner at any cost.

The two fictional sources

Source A: Cedar Hall venue sheet, version 1. Capacity: 24 seated. Hire: £120 for a three-hour session. Step-free entrance stated. Availability for the required date not supplied. No information about accessible toilets or the internal route to the room.

Source B: Rowan Rooms venue sheet, version 1. Capacity: 18 seated, 25 standing. Hire: from £90; duration and additional charges not supplied. Entrance access information not supplied. Availability for the required date not supplied.

Do not browse for these venues: they are invented for the exercise. The two paragraphs are the entire evidence set.

Give a bounded instruction

You can use this instruction with an approved tool:

Compare Cedar Hall and Rowan Rooms for a seated workshop of 20 people. Use only the two supplied source sheets. Separate confirmed facts from missing information. Do not infer availability, full accessibility or the final hire cost. Finish with questions the organiser must resolve before booking.

Keep the instruction and source versions beside the answer. If you edit either source, record the change. Otherwise you cannot tell whether a different answer resulted from a better instruction or different evidence.

Check this deliberately flawed draft

Rowan Rooms is the best choice. It holds 25 people, is fully accessible and costs £90 for the whole workshop. Both venues are available, so book Rowan to save £30.

Pause before reading the explanation. Underline each claim the sources do not support. Then write a corrected recommendation in fewer than 120 words.

What the draft gets wrong

Rowan's 25-person capacity is standing capacity. Its stated seated capacity is 18, below the requirement. The source supplies no entrance-access information, so “fully accessible” is unsupported. Even Cedar's step-free entrance would not establish every attendee's access needs.

Rowan's price is “from £90”, with no confirmed duration or additional charges. Subtracting it from Cedar's three-hour price does not establish a like-for-like saving. Neither source confirms availability.

A defensible conclusion is: Cedar meets the stated seated-capacity requirement, while Rowan does not on the information supplied. Cedar remains a candidate pending date availability, relevant accessibility requirements and confirmation that the session terms meet the organiser's needs. Neither venue is ready to book.

Notice that the corrected answer does not need to be longer or more impressive. It needs to preserve the conditions that change the decision.

Make a short evidence record

For each important statement, record the source and the supporting words. For missing facts, write “not supplied”. Keep three categories separate:

• Confirmed: Cedar's stated seated capacity is 24.
• Unsupported: Rowan is fully accessible.
• Unresolved: availability on the workshop date.

This record makes checking concrete. A link or citation is not enough if it points to a source that says something different. The language-model guide explains why a model can overlook an exception even when the source is present.

Repeat with one change

Change the group size to 18 and run the task again, or review your answer manually. Rowan now meets the stated seated-capacity requirement. Its price, accessibility and availability uncertainties remain.

A process that keeps repeating “Rowan is too small” has failed to apply the changed input. A process that now declares Rowan the winner has confused one resolved condition with a complete decision.

Record generation, review and correction time separately if you use a tool. One classroom exercise cannot establish a general productivity saving, but it can show where your effort went.

What to keep

Save the task, both sources, your instruction, the original output, corrections and final decision. If you only reviewed the supplied draft, label your record “supplied-output review”. Do not describe it as a live model test.

This is one step in a beginner-to-advanced AI learning path. Ampliflow AI Edge develops the same discipline across documents, data and supervised workflows: do the work, retain the evidence and know when the next action needs a person.