MEST Africa · AI for SMEs: 90-Day Plan Generator

This web app goes with the MEST Africa "AI for SMEs" Workshop (6 November 2026). SME owners fill in a 5-step form about their business, their pain points and their workshop takeaways. A Claude-powered planning agent then writes a personalised 90-day AI adoption plan. The app turns the plan into a branded PDF and emails it to the owner. The owner can also download it on the spot.

docs/sample-plan-mock-data.pdf shows the PDF layout, filled with placeholder data.

How it works

Form (5 steps)  →  POST /api/plan
                   ├─ Input guardrails: schema validation, sanitising, honeypot, rate limits, prompt-injection screen
                   ├─ Planning agent (Claude Opus 5.5, structured JSON output)
                   │    · system prompt built from the concept note: Modules 1–5, partners Oze / Kudobuzz / Intelli
                   │    · respects the owner's budget, weekly hours, devices and AI familiarity
                   │    · scope check: rejects non-genuine or harmful businesses and injection attempts
                   ├─ Output guardrails: 3 phases covering weeks 1–13, 3–5 processes, KPIs, safeguards,
                   │    and a ban on guaranteed results, fake reviews, scraped contact lists, pirated software, etc.
                   │    → one self-correction retry, then fail closed
                   ├─ PDF (@react-pdf/renderer, MEST branding)
                   ├─ Email with the PDF attached (Gmail/SMTP or Resend), optional BCC to a MEST inbox
                   └─ Optional lead webhook (Google Sheet / Airtable via Zapier, Make or Apps Script)

The form follows the workshop agenda:

1. About you: business, sector, location, team size, how you attended
2. Starting point: AI familiarity, channels, current tools, devices, budget, time available
3. Pain points: back-office (Module 2), marketing (Module 3), customer engagement (Module 4)
4. Takeaways: the 3 processes to automate (the workshop objective), most useful sessions, concerns
5. Goals: main 90-day goal, what success looks like, consent

Each PDF contains: a summary, strengths and opportunities, quick wins for this week, 3–5 priority processes with tools and costs, a week-by-week roadmap for weeks 1–13 in 3 phases, KPIs, a budget in GHS, responsible-AI safeguards (including Ghana's Data Protection Act, 2012 (Act 843)), skills to build, and next steps with MEST.

Deploy to Vercel

1. Push this folder to a GitHub repository.

2. In Vercel, choose Add New → Project, import the repository, and keep the defaults (Next.js is detected automatically).

3. Add these Environment Variables:

   Variable	Required	Notes
   ANTHROPIC_API_KEY	✅	From console.anthropic.com
   SMTP_USER / SMTP_PASS	email, option A	A Gmail / Google Workspace address and its app password. No DNS changes needed.
   RESEND_API_KEY	email, option B	From resend.com. The sending domain must be verified in Resend.
   EMAIL_FROM	✅ for email	e.g. MEST Africa <[email protected]>. With Gmail it must be SMTP_USER or one of its aliases.
   EMAIL_BCC	optional	MEST inbox that gets a copy of every plan
   LEADS_WEBHOOK_URL	optional	Receives each submission as JSON. Use it to build the SME lead list.
   RATE_LIMIT_PER_HOUR	optional	Requests per IP per hour (default 5)
   GOOGLE_SERVICE_ACCOUNT_EMAIL / GOOGLE_PRIVATE_KEY	for certificates	See "Certificates" below
   CERT_SHEET_ID and/or CERT_FOLDER_ID	for certificates	Where to find each attendee's certificate

4. Click Deploy.

Notes:

• Writing a plan takes about 1–2 minutes. The API route sets maxDuration = 300, which needs Vercel Fluid Compute (on by default for new projects). On a plan with a lower limit, raise the function duration in Project Settings → Functions.
• Without RESEND_API_KEY the app still works: users see the "download PDF" option instead of email delivery.
• The rate limiter keeps its counts in each serverless instance's memory, so it is approximate. For strict limits on a high-traffic launch, move it to Vercel KV or Upstash Redis (lib/guardrails.ts).
• On 6 November many participants submit from the same venue Wi-Fi, so they share one IP. Before the event, raise RATE_LIMIT_PER_HOUR (for example to 200). The per-email limit of 3 per hour still applies.

Certificates

/certificate (the "Get certificate" button in the nav) lets attendees download their certificate by entering the email they registered with. Certificates stay in Google Drive and don't need to be shared publicly. The app reads them through a Google service account.

1. In Google Cloud Console, create a project, enable the Google Drive API and Google Sheets API, then create a service account and download its JSON key.
2. Share the certificates folder (and sheet, if used) with the service account's email as Viewer.
3. Set in Vercel: GOOGLE_SERVICE_ACCOUNT_EMAIL (client_email from the JSON) and GOOGLE_PRIVATE_KEY (private_key from the JSON).
4. Tell the app how to find each certificate (one or both):
   • Sheet (CERT_SHEET_ID, the ID in the sheet's URL): a header row with an Email column and a column with each certificate's Drive link, such as the "Merged Doc URL" column Autocrat writes. An optional Name column personalises the greeting.
   • Folder (CERT_FOLDER_ID, the ID in the folder's URL): each certificate's file name contains the attendee's email, e.g. [email protected].
5. Redeploy and check /api/health shows "certificates": true.

Google Docs and Slides certificates are converted to PDF automatically. Keep each certificate under about 4 MB (Vercel's response limit).

Local development

cp .env.example .env.local   # fill in ANTHROPIC_API_KEY (and Resend if you want to test email)
npm install
npm run dev                  # http://localhost:3000

Project layout

app/page.tsx              Landing page (light theme, images, scroll animations, CTA to /plan)
app/plan/page.tsx         Form page
app/certificate/page.tsx  Certificate download page (animated headline)
lib/certificates.ts       Google Drive / Sheets certificate lookup
components/PlanForm.tsx   5-step form, loading and result screens, PDF download
app/api/plan/route.ts     Pipeline: validate → agent → PDF → email → lead webhook
lib/schema.ts             Form options, form schema, AI output schema
lib/agent.ts              System prompt and Claude call (structured output, refusal fallback, self-correction)
lib/guardrails.ts         Input, prompt and output guardrails, rate limiting
lib/pdf.tsx               Branded PDF document
lib/email.ts              Email template and sending