Caleb: added ajv json schema validation.

apps/api/package.json

@@ -51,6 +51,7 @@
     "@nangohq/node": "^0.36.33",
     "@sentry/node": "^7.48.0",
     "@supabase/supabase-js": "^2.7.1",
+    "ajv": "^8.12.0",
     "async": "^3.2.5",
     "async-mutex": "^0.4.0",
     "axios": "^1.3.4",

apps/api/pnpm-lock.yaml

@@ -35,6 +35,9 @@ dependencies:
   '@supabase/supabase-js':
     specifier: ^2.7.1
     version: 2.39.7
+  ajv:
+    specifier: ^8.12.0
+    version: 8.12.0
   async:
     specifier: ^3.2.5
     version: 3.2.5
@@ -1820,6 +1823,15 @@ packages:
       humanize-ms: 1.2.1
     dev: false
 
+  /ajv@8.12.0:
+    resolution: {integrity: sha512-sRu1kpcO9yLtYxBKvqfTeh9KzZEwO3STyX1HT+4CaDzC6HpTGYhIhPIzj9XuKU7KYDwnaeh5hcOwjy1QuJzBPA==}
+    dependencies:
+      fast-deep-equal: 3.1.3
+      json-schema-traverse: 1.0.0
+      require-from-string: 2.0.2
+      uri-js: 4.4.1
+    dev: false
+
   /ansi-escapes@4.3.2:
     resolution: {integrity: sha512-gKXj5ALrKWQLsYG9jlTRmR/xKluxHV+Z9QEwNIgCfM1/uwPMCuzVVnh5mwTd+OuBZcwSIMbqssNWRm1lE51QaQ==}
     engines: {node: '>=8'}
@@ -2926,6 +2938,10 @@ packages:
       - supports-color
     dev: false
 
+  /fast-deep-equal@3.1.3:
+    resolution: {integrity: sha512-f3qQ9oQy9j2AhBe/H9VC91wLmKBCCU/gDOnKNAYG5hswO7BLKj09Hc5HYNz9cGI++xlpDCIgDaitVs03ATR84Q==}
+    dev: false
+
   /fast-fifo@1.3.2:
     resolution: {integrity: sha512-/d9sfos4yxzpwkDkuN7k2SqFKtYNmCTzgfEpz82x34IM9/zc8KGxQoXg1liNC/izpRM/MBdt44Nmx41ZWqk+FQ==}
     dev: false
@@ -3999,6 +4015,10 @@ packages:
     hasBin: true
     dev: false
 
+  /json-schema-traverse@1.0.0:
+    resolution: {integrity: sha512-NM8/P9n3XjXhIZn1lLhkFaACTOURQXjWhV4BA/RnOv8xvgqtqpAX9IO4mRQxSx1Rlo4tqzeqb0sOlruaOy3dug==}
+    dev: false
+
   /json5@2.2.3:
     resolution: {integrity: sha512-XmOWe7eyHYH14cLdVPoyg+GOH3rYX++KpzrylJwSW98t3Nk+U8XOl8FWKOgwtzdb8lXGf6zYwDUzeHMWfxasyg==}
     engines: {node: '>=6'}
@@ -5264,6 +5284,11 @@ packages:
     resolution: {integrity: sha512-fGxEI7+wsG9xrvdjsrlmL22OMTTiHRwAMroiEeMgq8gzoLC/PQr7RsRDSTLUg/bZAZtF+TVIkHc6/4RIKrui+Q==}
     engines: {node: '>=0.10.0'}
 
+  /require-from-string@2.0.2:
+    resolution: {integrity: sha512-Xf0nWe6RseziFMu+Ap9biiUbmplq6S9/p+7w7YXP/JBHhrUDDUhwa+vANyubuqfZWTveU//DYVGsDG7RKL/vEw==}
+    engines: {node: '>=0.10.0'}
+    dev: false
+
   /resolve-cwd@3.0.0:
     resolution: {integrity: sha512-OrZaX2Mb+rJCpH/6CpSqt9xFVpN++x01XnN2ie9g6P5/3xelLAkXWVADpdz1IHD/KFfEXyE6V0U01OQ3UO2rEg==}
     engines: {node: '>=8'}
@@ -5970,6 +5995,12 @@ packages:
       picocolors: 1.0.0
     dev: true
 
+  /uri-js@4.4.1:
+    resolution: {integrity: sha512-7rKUyy33Q1yc98pQ1DAmLtwX109F7TIfWlW1Ydo8Wl1ii1SeHieeh0HHfPeL2fMXK6z0s8ecKs9frCuLJvndBg==}
+    dependencies:
+      punycode: 2.3.1
+    dev: false
+
   /urlpattern-polyfill@10.0.0:
     resolution: {integrity: sha512-H/A06tKD7sS1O1X2SshBVeA5FLycRpjqiBeqGKmBwBDBy28EnRjORxTNe269KSSr5un5qyWi1iL61wLxpd+ZOg==}
     dev: false

apps/api/src/__tests__/e2e_withAuth/index.test.ts

@@ -285,6 +285,11 @@ describe("E2E Tests for API Routes", () => {
         });
 
 
+      // Ensure that the job was successfully created before proceeding with LLM extraction
+      expect(response.statusCode).toBe(200);
+
+    
+      
       // Assuming the LLM extraction object is available in the response body under `data.llm_extraction`
       let llmExtraction = response.body.data.llm_extraction;
 

apps/api/src/controllers/scrape.ts

@@ -7,13 +7,14 @@ import { RateLimiterMode } from "../types";
 import { logJob } from "../services/logging/log_job";
 import { Document } from "../lib/entities";
 import { isUrlBlocked } from "../scraper/WebScraper/utils/blocklist"; // Import the isUrlBlocked function
+import Ajv from 'ajv';
 
 export async function scrapeHelper(
   req: Request,
   team_id: string,
   crawlerOptions: any,
   pageOptions: any,
-  extractorOptions: any
+  extractorOptions: ExtractorOptions
 ): Promise<{
   success: boolean;
   error?: string;
@@ -29,6 +30,7 @@ export async function scrapeHelper(
     return { success: false, error: "Firecrawl currently does not support social media scraping due to policy restrictions. We're actively working on building support for it.", returnCode: 403 };
   }
 
+
   const a = new WebScraperDataProvider();
   await a.setOptions({
     mode: "single_urls",

apps/api/src/lib/LLM-extraction/index.ts

@@ -3,6 +3,8 @@ import OpenAI from 'openai'
 // import { LlamaModel } from 'node-llama-cpp'
 import { z } from 'zod'
 import { zodToJsonSchema } from 'zod-to-json-schema'
+import Ajv from 'ajv';
+const ajv = new Ajv(); // Initialize AJV for JSON schema validation
 
 import {
     ScraperCompletionResult,
@@ -22,20 +24,29 @@ export async function generateCompletions(
 
     const switchVariable = "openAI" // Placholder, want to think more about how we abstract the model provider
 
+
     const completions = await Promise.all(documents.map(async (document: Document) => {
         switch (switchVariable) {
             case "openAI":
                 const llm = new OpenAI();
-                return await generateOpenAICompletions({
+                const completionResult = await generateOpenAICompletions({
                     client: llm,
                     document: document,
                     schema: schema,
                     prompt: prompt
                 });
+                // Validate the JSON output against the schema using AJV
+                const validate = ajv.compile(schema);
+                if (!validate(completionResult.llm_extraction)) {
+                    throw new Error(`LLM extraction did not match the extraction schema you provided. This could be because of a model hallucination, or an Error on our side. Try adjusting your prompt, and if it doesn't work reach out to support. AJV error: ${validate.errors?.map(err => err.message).join(', ')}`);
+                }
+
+                return completionResult;
             default:
                 throw new Error('Invalid client');
         }
     }));
+    
 
     return completions;
 }

apps/api/src/lib/LLM-extraction/models.ts

@@ -31,7 +31,7 @@ function prepareOpenAIDoc(
   return [{ type: 'text', text: document.markdown }]
 }
 
-export async function generateOpenAICompletions<T>({
+export async function generateOpenAICompletions({
   client,
   model = 'gpt-3.5-turbo',
   document,

apps/api/src/lib/entities.ts

@@ -57,7 +57,7 @@ export class Document {
   url?: string; // Used only in /search for now
   content: string;
   markdown?: string;
-  llm_extraction?: string;
+  llm_extraction?: Record<string, any>;
   createdAt?: Date;
   updatedAt?: Date;
   type?: string;