Franchise partner desk

Upload the case's Excel file (Applications + Territories, with the fixed Existing stores sheet). Every applicant gets a recommendation (Recommended, Waitlist, On hold or Rejected), the reason, next steps, a training plan and an email draft. A chat panel answers "why" questions and takes new facts ("APP-012 has now found a ground-floor space"), then re-runs the whole batch and reports every knock-on change.

No login. The file is read in the browser.

Run it locally

npm install
cp .env.example .env.local     # add OPENROUTER_API_KEY (or GEMINI_API_KEY)
npm run dev                    # http://localhost:3000

The app works without a key too: the chat falls back to built-in handling of simple questions and updates.

Deploy to Vercel (about 5 minutes)

1. Push this folder to a GitHub repo.
2. In Vercel: Add New → Project, import the repo. Framework is detected as Next.js.
3. Under Environment Variables, add OPENROUTER_API_KEY (and optionally OPENROUTER_MODEL, default google/gemini-3.7-flash). Or use a Google AI Studio key: GEMINI_API_KEY and GEMINI_MODEL. If both keys are set, OpenRouter is used.
4. Deploy. Open the link in a private window to confirm it works without being signed in.
5. Set a credit limit on the key (OpenRouter) or a spending cap (Google AI Studio). The link is public.

If the model ever returns errors (for example, a retired model name), set OPENROUTER_MODEL (or GEMINI_MODEL) to a current Flash model and redeploy. The dashboard keeps working either way.

How decisions are made

Code decides; Gemini only explains and turns chat messages into data updates. Same input, same output, every time.

1. Reject (hard rules): pin code already has an existing store; pin code isn't in any territory; investment more than 25% below the tier minimum.
2. Hold (fixable): missing required fields; city and pin code disagree; no space yet; upper or unknown floor; space under 300 sq ft; investment up to 25% short; form contradicts itself or the notes.
3. Score out of 100 for everyone else: investment headroom 30, experience 20, owned/rented space 25/20, ground/mall floor 15/12, launch timing 10.
4. Allocation per territory: highest score first, one store per pin code, up to the open slots, minimum score by territory priority (High 50, Medium 60, Low 70). The rest go to Waitlist.

Minimum investment: Tier 1 ₹30 lakh, Tier 2 ₹22 lakh, Tier 3 ₹18 lakh. All thresholds are editable under "Brand rules" in the app.

The sample file

public/sample-applications.xlsx is built on the real template and plants the traps judges are likely to use:

Applicant	Trap	Expected
APP-005	Pin 570009 already has ST-06	Rejected
APP-016	Pin outside every territory	Rejected
APP-009	₹18 lakh in Tier 1	Rejected
APP-008	₹26 lakh in Tier 1 (small shortfall)	On hold
APP-007	No space yet, 2-month launch	On hold
APP-012	Upper floor	On hold → Recommended after "found a ground-floor space", which pushes APP-013 off pin 572101
APP-015	City Mangaluru, pin from Mysuru	On hold
APP-017	"Not yet identified" but gives size and floor	On hold
APP-018	Investment blank	On hold
APP-019	Form says Ground, notes say second floor	On hold
APP-020	260 sq ft	On hold
APP-004, APP-010	Lose slots to higher scorers	Waitlist
APP-014	Score below a Low-priority territory's bar	Waitlist

Check it

npm run check   # 48 checks against the sample file

Files

• docs/: the original case PDF and template, plus HANDOFF.md

• scripts/check-sample.mjs: the self-check

• lib/parse.js: reads the template (skips the grey example rows, keeps pin codes as text)

• lib/engine.js: rules, scoring, allocation, before/after comparison

• lib/plan.js: next steps, training plans, email drafts

• lib/snapshot.js, app/api/chat/route.js: what Gemini sees, and the server call (key never reaches the browser)

• lib/chatLocal.js: chat fallback without Gemini

• app/page.js, app/components/*: the dashboard