MAPLE MCP server creation for RAG workflows

firestore.indexes.json

@@ -18,6 +18,40 @@
         }
       ]
     },
+    {
+      "collectionGroup": "publishedTestimony",
+      "queryScope": "COLLECTION_GROUP",
+      "fields": [
+        {
+          "fieldPath": "billId",
+          "order": "ASCENDING"
+        },
+        {
+          "fieldPath": "vector_embedding",
+          "vectorConfig": {
+            "dimension": 3072,
+            "flat": {}
+          }
+        }
+      ]
+    },
+    {
+      "collectionGroup": "publishedTestimony",
+      "queryScope": "COLLECTION_GROUP",
+      "fields": [
+        {
+          "fieldPath": "ballotQuestionId",
+          "order": "ASCENDING"
+        },
+        {
+          "fieldPath": "vector_embedding",
+          "vectorConfig": {
+            "dimension": 3072,
+            "flat": {}
+          }
+        }
+      ]
+    },
     {
       "collectionGroup": "archivedTestimony",
       "queryScope": "COLLECTION",
@@ -923,6 +957,45 @@
           "queryScope": "COLLECTION_GROUP"
         }
       ]
+    },
+    {
+      "collectionGroup": "bills",
+      "fieldPath": "vector_embedding",
+      "indexes": [
+        {
+          "queryScope": "COLLECTION_GROUP",
+          "vectorConfig": {
+            "dimension": 3072,
+            "flat": {}
+          }
+        }
+      ]
+    },
+    {
+      "collectionGroup": "publishedTestimony",
+      "fieldPath": "vector_embedding",
+      "indexes": [
+        {
+          "queryScope": "COLLECTION_GROUP",
+          "vectorConfig": {
+            "dimension": 3072,
+            "flat": {}
+          }
+        }
+      ]
+    },
+    {
+      "collectionGroup": "ballotQuestions",
+      "fieldPath": "vector_embedding",
+      "indexes": [
+        {
+          "queryScope": "COLLECTION",
+          "vectorConfig": {
+            "dimension": 3072,
+            "flat": {}
+          }
+        }
+      ]
     }
   ]
 }

functions/package.json

@@ -18,7 +18,8 @@
     "assemblyai": "^4.9.0",
     "axios": "^0.25.0",
     "date-fns": "^2.30.0",
-    "firebase-admin": "^11.11.1",
+    "firebase-admin": "^12.0.0",
+    "@google-cloud/aiplatform": "^3.9.0",
     "firebase-functions": "^5.1.1",
     "fluent-ffmpeg": "^2.1.3",
     "fuse.js": "6.5.3",

functions/src/ballotQuestions/vector.ts

@@ -0,0 +1,8 @@
+import { createVectorIndexer } from "../search/createVectorIndexer"
+
+export const syncBallotQuestionToVectorIndex = createVectorIndexer({
+  documentTrigger: "ballotQuestions/{id}",
+  textFields: ["title", "description", "fullSummary"],
+  vectorField: "vector_embedding",
+  titleField: "title"
+})

functions/src/bills/vector.ts

@@ -0,0 +1,8 @@
+import { createVectorIndexer } from "../search/createVectorIndexer"
+
+export const syncBillToVectorIndex = createVectorIndexer({
+  documentTrigger: "generalCourts/{court}/bills/{id}",
+  textFields: ["content.Title", "content.DocumentText"],
+  vectorField: "vector_embedding",
+  titleField: "content.Title"
+})

functions/src/search/createVectorIndexer.ts

@@ -0,0 +1,98 @@
+import { runWith } from "firebase-functions"
+import * as admin from "firebase-admin"
+import { PredictionServiceClient, helpers } from "@google-cloud/aiplatform"
+import hash from "object-hash"
+
+export interface VectorIndexerConfig {
+  documentTrigger: string
+  textFields: string[] // Fields to combine for the embedding
+  vectorField: string // Destination field for the embedding (e.g., 'vector_embedding')
+  titleField?: string // Optional field to use as the title for prefixing
+}
+
+export function createVectorIndexer(config: VectorIndexerConfig) {
+  const location = "us-central1"
+  const publisher = "google"
+  const model = "gemini-embedding-2"
+
+  return runWith({
+    timeoutSeconds: 60,
+    memory: "512MB",
+    secrets: ["GOOGLE_APPLICATION_CREDENTIALS"] // If needed, though usually automatic on GCP
+  })
+    .firestore.document(config.documentTrigger)
+    .onWrite(async change => {
+      const data = change.after.exists ? change.after.data() : null
+      if (!data) return // Deleted
+
+      // Extract text to embed
+      const textToEmbed = config.textFields
+        .map(field => {
+          const parts = field.split(".")
+          let val: any = data
+          for (const part of parts) val = val?.[part]
+          return val
+        })
+        .filter(Boolean)
+        .join("\n\n")
+
+      if (!textToEmbed) return
+
+      // Extract title for gemini-embedding-2 prefixing
+      let title = "none"
+      if (config.titleField) {
+        const parts = config.titleField.split(".")
+        let val: any = data
+        for (const part of parts) val = val?.[part]
+        title = val || "none"
+      }
+
+      // Check if text has changed to avoid redundant API calls
+      const textHash = hash({ textToEmbed, title })
+      const previousHash = (
+        change.before.exists ? change.before.data() : null
+      )?.[`${config.vectorField}_hash`]
+
+      if (textHash === previousHash && data[config.vectorField]) {
+        return // Nothing changed
+      }
+
+      // Initialize Vertex AI client
+      const project = admin.app().options.projectId
+      const endpoint = `projects/${project}/locations/${location}/publishers/${publisher}/models/${model}`
+      const client = new PredictionServiceClient({
+        apiEndpoint: `${location}-aiplatform.googleapis.com`
+      })
+
+      // Get embedding with multimodal/task prefix
+      const formattedText = `title: ${title} | text: ${textToEmbed}`
+      const instance = helpers.toValue({ content: formattedText })!
+      const responseArray = (await client.predict({
+        endpoint,
+        instances: [instance]
+      })) as any
+      const response = responseArray[0]
+
+      if (!response.predictions || response.predictions.length === 0) {
+        throw new Error("No predictions returned from Vertex AI")
+      }
+
+      const prediction = helpers.fromValue(
+        response.predictions[0] as any
+      ) as any
+      const embedding =
+        prediction.embeddings?.values || prediction.embedding?.values
+
+      if (!embedding) {
+        throw new Error(
+          `Unexpected prediction format: ${JSON.stringify(prediction)}`
+        )
+      }
+
+      // Update document
+      await change.after.ref.update({
+        [config.vectorField]: embedding,
+        [`${config.vectorField}_hash`]: textHash // Store hash to track changes
+      })
+    })
+}

functions/src/testimony/vector.ts

@@ -0,0 +1,8 @@
+import { createVectorIndexer } from "../search/createVectorIndexer"
+
+export const syncTestimonyToVectorIndex = createVectorIndexer({
+  documentTrigger: "users/{uid}/publishedTestimony/{id}",
+  textFields: ["content"],
+  vectorField: "vector_embedding",
+  titleField: "billTitle"
+})