Donation intake and inventory intelligence system for WellSpring Women's Center, Sacramento.

• Intake — staff scan barcodes or manually enter donated items on their phone
• Classification — Gemini 2.5 Flash automatically sorts items into categories
• Review — low-confidence items go to a human review queue
• Dashboard — category breakdowns and weekly trends
• Give Out — record items distributed to guests
• Ask — natural-language questions answered from live inventory data

VITE_SUPABASE_URL=https://<project>.supabase.co
VITE_SUPABASE_ANON_KEY=<anon key>
VITE_API_BASE_URL=https://<api-gateway-id>.execute-api.us-west-2.amazonaws.com/dev

SUPABASE_URL
SUPABASE_SERVICE_ROLE_KEY
GEMINI_API_KEY
OLLAMA_HOST   (default: http://localhost:11434)

pnpm install
pnpm --filter web dev        # frontend on http://localhost:5173

Run these in PowerShell from the repo root every time you change anything in apps/api/:

$env:SUPABASE_URL = "https://<project>.supabase.co"
$env:SUPABASE_SERVICE_ROLE_KEY = "<service role key>"
$env:GEMINI_API_KEY = "<gemini key>"
$env:OLLAMA_HOST = "http://localhost:11434"

cd apps/api
pnpm build
serverless deploy

pnpm build runs esbuild and outputs a single dist/index.js. serverless deploy packages and uploads it to AWS Lambda.

After a successful deploy, Serverless prints the new API Gateway URL. If it changed, update VITE_API_BASE_URL in apps/web/.env and redeploy the frontend.

Run this from the repo root every time you change anything in apps/web/:

cd C:\Users\USER\Documents\GitHub\HerLedger
pnpm --filter web build
wrangler pages deploy apps/web/dist --project-name herledger --commit-dirty=true

pnpm --filter web build compiles the React app into apps/web/dist/. wrangler pages deploy uploads it to Cloudflare Pages.

Migrations live in packages/db/migrations/. They are numbered and applied in order. Never edit an already-applied migration — create a new one instead.

To apply a new migration, paste its contents into the Supabase SQL Editor and run it.

Current migrations:

Seed data (run once after initial setup):

packages/db/seed/categories.sql   — inserts all 30 item categories

1. Staff scan items in Intake — each item is saved as pending
2. Press Run Classification on the Dashboard
3. Lambda calls Gemini 2.5 Flash for each pending item
4. Items classified with confidence ≥ 75% → classified
5. Items below threshold → needs_review (appear in Review queue)
6. After all items are processed, refresh_daily_snapshots() updates the dashboard

Gemini free tier: 20 requests/day for gemini-2.5-flash. Enable billing at aistudio.google.com for production use.

"0 items classified" after pressing Run Classification Items are not in pending status. Run in Supabase SQL Editor:

UPDATE items SET status = 'pending'
WHERE id NOT IN (SELECT DISTINCT item_id FROM item_classifications);

Classification keeps looping / never finishes Items stuck as needs_review with no classification record. Same fix as above.

Dashboard shows wrong counts Refresh the snapshot manually:

SELECT refresh_daily_snapshots();

429 Too Many Requests from Gemini Free tier quota exhausted (20 req/day). Wait until tomorrow or enable billing.

API returns 500 Check AWS CloudWatch: Console → CloudWatch → Log groups → /aws/lambda/herledger-api-dev-api → latest log stream.