Add OCR rate limit handling and batch reprocess action

Rate limiting:
- Retry with exponential backoff (5s, 10s, 20s, 40s) on rate limit errors
- Reduce concurrent OCR jobs from 2 to 1
- Add 2s delay between AI calls within a card (response vs survey side)
- Add 3s delay between cards in a multi-page PDF
- Add 3s delay between jobs in the processing queue

Batch reprocess:
- New /api/cards/reprocess-batch endpoint processes cards sequentially
  with delays to respect rate limits
- Reprocess button added to selection toolbar on dashboard

Made-with: Cursor

src/app/api/cards/reprocess-batch/route.ts

@@ -0,0 +1,62 @@
+import { NextRequest, NextResponse } from "next/server";
+import { prisma } from "@/lib/db";
+import { reprocessCard } from "@/lib/ocr";
+
+const DELAY_BETWEEN_CARDS_MS = 3_000;
+
+export async function POST(request: NextRequest) {
+  try {
+    const body = await request.json().catch(() => ({}));
+    const ids = body.ids as string[] | undefined;
+
+    if (!ids || !Array.isArray(ids) || ids.length === 0) {
+      return NextResponse.json({ error: "No card IDs provided" }, { status: 400 });
+    }
+
+    const cards = await prisma.responseCard.findMany({
+      where: { id: { in: ids } },
+      select: { id: true, ocrStatus: true, backImagePath: true, frontImagePath: true },
+    });
+
+    const eligible = cards.filter(
+      (c) => c.ocrStatus !== "processing" && (c.backImagePath || c.frontImagePath)
+    );
+
+    if (eligible.length === 0) {
+      return NextResponse.json(
+        { error: "No eligible cards (already processing or no images)" },
+        { status: 400 }
+      );
+    }
+
+    await prisma.responseCard.updateMany({
+      where: { id: { in: eligible.map((c) => c.id) } },
+      data: { ocrStatus: "processing", ocrError: null },
+    });
+
+    (async () => {
+      for (let i = 0; i < eligible.length; i++) {
+        try {
+          await reprocessCard(eligible[i].id);
+        } catch (err) {
+          console.error(`[reprocess-batch] Card ${eligible[i].id} failed:`, err);
+        }
+        if (i < eligible.length - 1) {
+          await new Promise((r) => setTimeout(r, DELAY_BETWEEN_CARDS_MS));
+        }
+      }
+      console.log(`[reprocess-batch] Completed ${eligible.length} card(s)`);
+    })().catch((err) => {
+      console.error("[reprocess-batch] Background task failed:", err);
+    });
+
+    return NextResponse.json({
+      ok: true,
+      queued: eligible.length,
+      skipped: ids.length - eligible.length,
+    });
+  } catch (error) {
+    console.error("[reprocess-batch POST]", error);
+    return NextResponse.json({ error: "Failed to start batch reprocessing" }, { status: 500 });
+  }
+}

src/components/cards/dashboard-content.tsx

@@ -145,6 +145,28 @@ export function DashboardContent() {
     }
   };
 
+  const handleBulkReprocess = async (ids: string[]) => {
+    try {
+      const res = await fetch("/api/cards/reprocess-batch", {
+        method: "POST",
+        headers: { "Content-Type": "application/json" },
+        body: JSON.stringify({ ids }),
+      });
+      const data = await res.json();
+      if (res.ok) {
+        toast.success(
+          `Reprocessing ${data.queued} card(s)${data.skipped > 0 ? ` (${data.skipped} skipped)` : ""}`
+        );
+        setSelectedIds([]);
+        fetchCards();
+      } else {
+        toast.error(data.error || "Batch reprocess failed");
+      }
+    } catch {
+      toast.error("Failed to start batch reprocessing");
+    }
+  };
+
   const handleExportCsv = () => {
     if (data.length === 0) {
       toast.error("No data to export");
@@ -344,6 +366,7 @@ export function DashboardContent() {
         selectedIds={selectedIds}
         onMarkReviewed={(ids) => handleBulkAction(ids, "reviewed")}
         onMarkExported={(ids) => handleBulkAction(ids, "exported")}
+        onReprocess={handleBulkReprocess}
         onDelete={(ids) => handleBulkAction(ids, "delete")}
         onClear={() => setSelectedIds([])}
       />

src/components/cards/selection-toolbar.tsx

@@ -3,6 +3,7 @@
 import {
   CheckCircle,
   Download,
+  RefreshCw,
   Trash2,
   X,
 } from "lucide-react";
@@ -13,6 +14,7 @@ interface SelectionToolbarProps {
   selectedIds: string[];
   onMarkReviewed?: (ids: string[]) => void;
   onMarkExported?: (ids: string[]) => void;
+  onReprocess?: (ids: string[]) => void;
   onDelete?: (ids: string[]) => void;
   onClear: () => void;
 }
@@ -21,6 +23,7 @@ export function SelectionToolbar({
   selectedIds,
   onMarkReviewed,
   onMarkExported,
+  onReprocess,
   onDelete,
   onClear,
 }: SelectionToolbarProps) {
@@ -63,6 +66,17 @@ export function SelectionToolbar({
               <span className="hidden sm:inline ml-1">Export</span>
             </Button>
           )}
+          {onReprocess && (
+            <Button
+              variant="ghost"
+              size="sm"
+              className="rounded-xl"
+              onClick={() => onReprocess(selectedIds)}
+            >
+              <RefreshCw className="size-4" />
+              <span className="hidden sm:inline ml-1">Reprocess</span>
+            </Button>
+          )}
           {onDelete && (
             <Button
               variant="ghost"

src/lib/ai-ocr.ts

@@ -99,32 +99,53 @@ export interface OcrResult {
   side: "response" | "survey";
 }
 
+const MAX_RETRIES = 4;
+const INITIAL_BACKOFF_MS = 5_000;
+
 async function extractStructured<T extends Record<string, unknown>>(
   model: LanguageModel,
   schema: z.ZodType<T>,
   systemPrompt: string,
   imageBase64: string
 ): Promise<T> {
-  const { output, text } = await generateText({
-    model,
-    output: Output.object({ schema }),
-    messages: [
-      { role: "system", content: systemPrompt },
-      {
-        role: "user",
-        content: [
-          { type: "text", text: "Extract all data from this scanned card image." },
-          { type: "image", image: Buffer.from(imageBase64, "base64") },
-        ],
-      },
-    ],
-  });
+  let lastError: unknown;
 
-  if (!output) {
-    throw new Error(`AI model did not return structured output. Raw text: ${(text || "").slice(0, 500)}`);
+  for (let attempt = 0; attempt <= MAX_RETRIES; attempt++) {
+    try {
+      const { output, text } = await generateText({
+        model,
+        output: Output.object({ schema }),
+        messages: [
+          { role: "system", content: systemPrompt },
+          {
+            role: "user",
+            content: [
+              { type: "text", text: "Extract all data from this scanned card image." },
+              { type: "image", image: Buffer.from(imageBase64, "base64") },
+            ],
+          },
+        ],
+      });
+
+      if (!output) {
+        throw new Error(`AI model did not return structured output. Raw text: ${(text || "").slice(0, 500)}`);
+      }
+
+      return output;
+    } catch (err) {
+      lastError = err;
+      const msg = err instanceof Error ? err.message : String(err);
+      const isRateLimit = /rate.?limit|too many requests|429|temporarily|free credits/i.test(msg);
+
+      if (!isRateLimit || attempt === MAX_RETRIES) throw err;
+
+      const backoff = INITIAL_BACKOFF_MS * Math.pow(2, attempt);
+      console.log(`[ai-ocr] Rate limited (attempt ${attempt + 1}/${MAX_RETRIES + 1}), retrying in ${backoff / 1000}s...`);
+      await new Promise((r) => setTimeout(r, backoff));
+    }
   }
 
-  return output;
+  throw lastError;
 }
 
 export async function ocrImage(

src/lib/ocr.ts

@@ -5,6 +5,11 @@ import { ocrImage } from "./ai-ocr";
 import { pdfToImages, imageToBase64, processUploadedImage } from "./pdf";
 import { fireIntegrationEvent } from "./integrations";
 
+const DELAY_BETWEEN_OCR_CALLS_MS = 2_000;
+function ocrDelay() {
+  return new Promise((r) => setTimeout(r, DELAY_BETWEEN_OCR_CALLS_MS));
+}
+
 export async function processFile(
   jobId: string,
   fileName: string,
@@ -54,7 +59,9 @@ export async function processFile(
       pairs.push({ response: pageImages[0] });
     }
 
-    for (const pair of pairs) {
+    for (let pairIdx = 0; pairIdx < pairs.length; pairIdx++) {
+      if (pairIdx > 0) await ocrDelay();
+      const pair = pairs[pairIdx];
       const card = await prisma.responseCard.create({
         data: {
           sourceFile: sourceKey,
@@ -85,6 +92,8 @@ export async function processFile(
         }
 
         if (pair.survey) {
+          if (pair.response) await ocrDelay();
+
           const imgKey = `images/${card.id}/survey.jpg`;
           await uploadBuffer(imgKey, pair.survey.buffer, "image/jpeg");
           await prisma.responseCard.update({
@@ -196,6 +205,8 @@ export async function reprocessCard(cardId: string): Promise<void> {
     }
 
     if (card.frontImagePath) {
+      if (card.backImagePath) await ocrDelay();
+
       const imgBuffer = await getBuffer(card.frontImagePath);
       const base64 = await imageToBase64(imgBuffer);
       const result = await ocrImage(base64, "survey");

src/lib/processing-queue.ts

@@ -1,6 +1,7 @@
 import { processFile } from "./ocr";
 
-const MAX_CONCURRENT = 2;
+const MAX_CONCURRENT = 1;
+const DELAY_BETWEEN_JOBS_MS = 3_000;
 const queue: QueueItem[] = [];
 let running = 0;
 
@@ -22,7 +23,9 @@ function drain() {
       .catch(item.reject)
       .finally(() => {
         running--;
-        drain();
+        if (queue.length > 0) {
+          setTimeout(drain, DELAY_BETWEEN_JOBS_MS);
+        }
       });
   }
 }
@@ -35,7 +38,7 @@ export function enqueueProcessing(
 ): Promise<string[]> {
   return new Promise((resolve, reject) => {
     queue.push({ jobId, fileName, fileBuffer, isPdf, resolve, reject });
-    drain();
+    if (running < MAX_CONCURRENT) drain();
   });
 }