Capstone: pick your AI use case

Three ways to see it

1. 1

   The template is five fields on one page. Problem: state in two lines what work you are doing now, and what hurts about it. Before: how long it takes, how often it goes wrong, who picks up the cost when it does. AI tools used: which model, which prompt, which other software in the chain. After: what the workflow looks like once AI is folded in. Measurement: the number you will look at in thirty days to decide if it worked.

2. 2

   An example, from a real cohort. Maria, Assistant Manager at a public-sector hospital in Rawalpindi, picked the problem of incoming OPD referral letters. Before AI: forty-five referrals a day, each read in full, average eight minutes per letter, six hours of her staff's time. AI in chain: Urdu OCR for hand-written letters, then Claude with a fixed prompt to extract patient name, referring doctor, suspected diagnosis, and urgency tier into a table. After: a junior nurse spot-checks the table against originals in twenty seconds per row. Total time, ninety minutes. Measurement: hours saved per week, plus error rate sampled at five percent.

3. 3

   Way one to measure: time saved. The cleanest number. Time a sample of ten tasks before, ten after. If the difference is under twenty percent, the workflow is not worth keeping, because the verification cost will eat the saving. If the difference is above fifty percent, you have a serious win and you should keep going.

The why-tree

Try this with Claude

AI-edge prompt: 'Here is my one-page capstone draft. Critique it like a sceptical board member. Where am I overestimating the benefit, underestimating the risk, or skipping a verification step? Suggest the single change that would make this plan more honest, not more ambitious.' The right answer makes the plan smaller, sharper, and likelier to succeed.