Capstone: draft a one-page AI RFP for your department

Three ways to see it

1. 1

   A working one-page RFP has seven blocks. Title and need: one sentence each. Scope: bullet list of what is in, what is out. Functional outcomes: three to five outcomes the system must produce. Non-functional: latency, languages (must include Urdu and at least one regional language), accessibility (WCAG 2.2 AA), security baseline. Inclusion mandates: female participation, PWD accessibility, local capacity thresholds. Evaluation matrix: five legs with weights. Commercial envelope: budget range and term.

2. 2

   Way one to write the title and need: be specific and small. 'AI chatbot for federal ombudsman to triage citizen complaints in Urdu and Sindhi, replacing the current paper-only intake at six regional offices.' That sentence rules in three things and rules out fifty. Vague titles produce vague bids.

3. 3

   Way two: pick three to five functional outcomes that can be measured. Triage 95 percent of incoming complaints to a category in under three seconds. Detect 90 percent of duplicate complaints across the six regions. Produce a daily report in Urdu and English for the ombudsman's office. Each outcome is testable with the data you already have, not a vendor claim that must be taken on faith.

The why-tree

Try this with Claude

AI-edge prompt to try: 'Acting as the head of a Pakistani federal IT division, draft a one-page AI RFP for the use case I describe. Include all seven blocks, lock the evaluation matrix to weights I supply, and mandate Urdu plus one regional language, female participation, and PWD accessibility as hard thresholds.' Edit before issuing.