JMCP: Papers - jmcp/use-structured-output-for-tag-enrichment-eu897mic

apps/web/trigger/tasks.ts

@@ -6,14 +6,9 @@ const provider = new GrokProvider()
 export const enrichPaperMetadata = task({
   id: "enrich-paper-metadata",
   run: async (payload: { title: string; abstract: string }) => {
-    const tags = await provider.complete(
-      "tag-extraction",
-      `Title: ${payload.title}\nAbstract: ${payload.abstract}`,
-    )
+    const result = await provider.extractTags(payload.title, payload.abstract)
 
-    return {
-      tags,
-    }
+    return result
   },
 })
 

packages/ai/src/index.ts

@@ -1,5 +1,12 @@
 import { getPapersConfig } from "@papers/config"
-import { type ProviderTaskKind, type SafeAiPayload, safeAiPayloadSchema } from "@papers/contracts"
+import {
+  type ProviderTaskKind,
+  type SafeAiPayload,
+  safeAiPayloadSchema,
+  type TagExtractionResult,
+  tagExtractionJsonSchema,
+  tagExtractionResultSchema,
+} from "@papers/contracts"
 
 export class GrokProvider {
   readonly #config = getPapersConfig()
@@ -61,4 +68,59 @@ export class GrokProvider {
     const payload = (await response.json()) as { output_text?: string }
     return payload.output_text ?? ""
   }
+
+  async extractTags(title: string, abstract: string): Promise<TagExtractionResult> {
+    const text = `Title: ${title}\nAbstract: ${abstract}`
+
+    this.assertSafe({
+      task: "tag-extraction",
+      isBlindContent: false,
+      containsPrivateDraft: false,
+      text,
+    })
+
+    if (!this.enabled || !this.#config.PAPERS_XAI_API_KEY) {
+      return { tags: [] }
+    }
+
+    const response = await fetch(`${this.#config.PAPERS_XAI_BASE_URL}/responses`, {
+      method: "POST",
+      headers: {
+        authorization: `Bearer ${this.#config.PAPERS_XAI_API_KEY}`,
+        "content-type": "application/json",
+      },
+      body: JSON.stringify({
+        model: this.#config.PAPERS_XAI_MODEL,
+        input: [
+          {
+            role: "system",
+            content: [
+              {
+                type: "input_text",
+                text: "Extract up to 8 research topic tags from the paper. Each tag needs a human-readable label and a URL-safe slug. Never infer hidden identity.",
+              },
+            ],
+          },
+          {
+            role: "user",
+            content: [{ type: "input_text", text }],
+          },
+        ],
+        text: {
+          format: {
+            type: "json_schema",
+            ...tagExtractionJsonSchema,
+          },
+        },
+      }),
+    })
+
+    if (!response.ok) {
+      throw new Error(`Grok tag extraction failed with ${response.status}`)
+    }
+
+    const raw = (await response.json()) as { output_text?: string }
+    const parsed: unknown = JSON.parse(raw.output_text ?? "{}")
+    return tagExtractionResultSchema.parse(parsed)
+  }
 }

packages/contracts/src/index.ts

@@ -181,6 +181,43 @@ export const safeAiPayloadSchema = z.object({
 })
 export type SafeAiPayload = z.infer<typeof safeAiPayloadSchema>
 
+export const tagExtractionResultSchema = z.object({
+  tags: z
+    .array(
+      z.object({
+        label: z.string(),
+        slug: z.string(),
+      }),
+    )
+    .max(8),
+})
+export type TagExtractionResult = z.infer<typeof tagExtractionResultSchema>
+
+/** JSON Schema sent to the Grok Responses API for structured output. */
+export const tagExtractionJsonSchema = {
+  name: "tag_extraction",
+  strict: true,
+  schema: {
+    type: "object",
+    properties: {
+      tags: {
+        type: "array",
+        items: {
+          type: "object",
+          properties: {
+            label: { type: "string" },
+            slug: { type: "string" },
+          },
+          required: ["label", "slug"],
+          additionalProperties: false,
+        },
+      },
+    },
+    required: ["tags"],
+    additionalProperties: false,
+  },
+} as const
+
 export const opportunityIdeaSchema = z.object({
   id: z.string(),
   label: z.string(),