feat: chat with pdf

packages/lib/server-only/context.ts

@@ -0,0 +1,35 @@
+import { Pinecone } from '@pinecone-database/pinecone';
+
+import { getEmbeddings } from './embeddings';
+
+export async function getMatchesFromEmbeddings(embeddings: number[]) {
+  const pc = new Pinecone({
+    apiKey: process.env.PINECONE_API_KEY!,
+    environment: process.env.PINECONE_ENV!,
+  });
+
+  const pineconeIndex = pc.index('documenso-chat-with-pdf-test');
+
+  try {
+    const queryResult = await pineconeIndex.query({
+      topK: 5,
+      vector: embeddings,
+      includeMetadata: true,
+    });
+
+    return queryResult.matches || [];
+  } catch (error) {
+    console.error('There was an error getting matches from embeddings: ', error);
+    throw new Error('There was an error getting matches from embeddings');
+  }
+}
+
+export async function getContext(query: string) {
+  const queryEmbeddings = await getEmbeddings(query);
+  const matches = await getMatchesFromEmbeddings(queryEmbeddings);
+
+  const qualifyingMatches = matches.filter((match) => match.score && match.score > 0.7);
+  const docs = qualifyingMatches.map((match) => match.metadata?.text);
+
+  return docs.join('\n').substring(0, 3000);
+}

packages/lib/server-only/embeddings.ts

@@ -0,0 +1,23 @@
+import { Configuration, OpenAIApi } from 'openai-edge';
+
+const config = new Configuration({
+  apiKey: process.env.OPENAI_API_KEY!,
+});
+
+const openai = new OpenAIApi(config);
+
+export async function getEmbeddings(text: string) {
+  try {
+    const response = await openai.createEmbedding({
+      model: 'text-embedding-ada-002',
+      input: text.replace(/\n/g, ' '),
+    });
+
+    const result = await response.json();
+
+    return result.data[0].embedding;
+  } catch (error) {
+    console.error('There was an error getting embeddings: ', error);
+    throw new Error('There was an error getting embeddings');
+  }
+}

packages/lib/server-only/pinecone.ts

@@ -0,0 +1,113 @@
+import { Pinecone } from '@pinecone-database/pinecone';
+import { PDFLoader } from 'langchain/document_loaders/fs/pdf';
+import { RecursiveCharacterTextSplitter } from 'langchain/text_splitter';
+import md5 from 'md5';
+
+import { getEmbeddings } from './embeddings';
+
+let pc: Pinecone | null = null;
+
+// export type PDFPage = {
+//   pageContent: string;
+//   metadata: {
+//     source: string;
+//     pdf: {
+//       version: string;
+//       info: {
+//         pdfformatversion: string;
+//         isacroformpresent: boolean;
+//         isxfapresent: boolean;
+//         creator: string;
+//         producer: string;
+//         ceationdate: string;
+//         moddate: string;
+//       };
+//       metadata: null;
+//       totalPages: number;
+//     };
+//     loc: {
+//       pageNumber: number;
+//     };
+//   };
+// };
+
+export type PDFPage = unknown;
+export const getPineconeClient = () => {
+  if (!pc) {
+    pc = new Pinecone({
+      apiKey: process.env.PINECONE_API_KEY!,
+      environment: process.env.PINECONE_ENV!,
+    });
+  }
+
+  return pc;
+};
+
+export async function loadFileIntoPinecone(file: string) {
+  if (!file) {
+    throw new Error('No file provided');
+  }
+
+  const loader = new PDFLoader(file);
+  const pages: PDFPage[] = await loader.load();
+
+  const documents = await Promise.all(pages.map(prepareDocument));
+
+  const vectors = await Promise.all(documents.flat().map(embedDocuments));
+
+  const client = getPineconeClient();
+  const pineconeIndex = client.index('documenso-chat-with-pdf-test');
+
+  try {
+    await pineconeIndex.upsert(vectors);
+  } catch (error) {
+    console.error('There was an error upserting vectors: ', error);
+  }
+}
+
+async function embedDocuments(doc) {
+  try {
+    const embeddings = await getEmbeddings(doc.pageContent);
+    const hash = md5(doc.pageContent);
+
+    return {
+      id: hash,
+      values: embeddings,
+      metadata: {
+        text: doc.metadata.text,
+        pageNumber: doc.metadata.pageNumber,
+      },
+    };
+  } catch (error) {
+    console.error('There was an error embedding documents: ', error);
+    throw new Error('There was an error embedding documents');
+  }
+}
+
+export const truncateStringByBytes = (str: string, numBytes: number) => {
+  const encoder = new TextEncoder();
+
+  return new TextDecoder('utf-8').decode(encoder.encode(str).slice(0, numBytes));
+};
+
+async function prepareDocument(page: PDFPage) {
+  let { pageContent, metadata } = page;
+  pageContent = pageContent.replace(/\n/g, '');
+
+  const splitter = new RecursiveCharacterTextSplitter();
+  const docs = await splitter.splitDocuments([
+    {
+      pageContent,
+      metadata: {
+        pageNumber: metadata.loc.pageNumber,
+        text: truncateStringByBytes(pageContent, 36000),
+      },
+    },
+  ]);
+
+  return docs;
+}
+
+function convertToAscii(input: string) {
+  return input.replace(/[^\x00-\x7F]/g, '');
+}