Embeddings

src/adapters/supabase/helpers/client.ts

@@ -573,3 +573,41 @@ export const _approveLabelChange = async (changeId: number) => {
 
   return;
 };
+
+/**
+ * Upserts embeddings vector to the `embeddings` table
+ * @param org - The organization name
+ * @param repo - The repository name
+ * @param issue - The issue number
+ * @param embeddings - The vector of floating point numbers
+ */
+export const upsertEmbeddings = async (org: string, repo: string, issue: number, embedding: number[]) => {
+  const { supabase } = getAdapters();
+  const logger = getLogger();
+  const { data } = await supabase.from("embeddings").select("*").eq("org", org).eq("repo", repo).eq("issue", issue).single();
+  if (data) {
+    // Update the existing record with the new embedding for a given set.
+    const id = data["id"] as number;
+    const { error } = await supabase.from("embeddings").upsert({
+      id: id,
+      embedding,
+      updated_at: new Date().toISOString(),
+    });
+    if (error) {
+      logger.info(`Updating embedding table failed. id: ${id}, org: ${org}, repo: ${repo}, issue: ${issue}, embedding: ${embedding.join(", ")}`);
+    }
+  } else {
+    // Insert a new record to the `embeddings` table.
+    const { error } = await supabase.from("embeddings").upsert({
+      org,
+      repo,
+      issue,
+      embedding,
+      created_at: new Date().toISOString(),
+      updated_at: new Date().toISOString(),
+    });
+    if (error) {
+      logger.info(`Inserting to embedding table failed. org: ${org}, repo: ${repo}, issue: ${issue}, embedding: ${embedding.join(", ")}`);
+    }
+  }
+};

src/handlers/comment/action.ts

@@ -51,11 +51,9 @@ export const handleComment = async (): Promise<void> => {
         const callbackComment = response ?? successComment ?? "";
         if (callbackComment) await callback(issue.number, callbackComment, payload.action, payload.comment);
       } catch (err: unknown) {
-        // Use failureComment for failed command if it is available
-        if (failureComment) {
-          await callback(issue.number, failureComment, payload.action, payload.comment);
+        for (const comment of [failureComment, ErrorDiff(err)]) {
+          if (comment) await callback(issue.number, comment, payload.action, payload.comment);
         }
-        await callback(issue.number, ErrorDiff(err), payload.action, payload.comment);
       }
     } else {
       logger.info(`Skipping for a command: ${command}`);

src/handlers/issue/embeddings.ts

@@ -0,0 +1,29 @@
+import { getBotContext, getLogger } from "../../bindings";
+import { Payload } from "../../types";
+import { generateEmbeddings } from "../../helpers";
+import { upsertEmbeddings } from "../../adapters/supabase";
+
+/**
+ * Generates an embedding vector for the current issue
+ */
+export const embeddings = async () => {
+  const { payload: _payload } = getBotContext();
+  const logger = getLogger();
+  const payload = _payload as Payload;
+  const issue = payload.issue;
+
+  if (!issue) {
+    logger.info(`Skip to generate embeddings because of no issue instance`);
+    return;
+  }
+
+  if (!issue.body) {
+    logger.info("Skip to generate embeddings because of empty body");
+    return;
+  }
+
+  const embeddings = await generateEmbeddings(issue.body);
+  if (embeddings.length > 0) {
+    await upsertEmbeddings(payload.repository.owner.login, payload.repository.name, issue.number, embeddings);
+  }
+};

src/handlers/issue/index.ts

@@ -1 +1,2 @@
 export * from "./pre";
+export * from "./embeddings";

src/handlers/processors.ts

@@ -7,14 +7,14 @@ import { handleComment, issueClosedCallback, issueCreatedCallback, issueReopened
 import { checkPullRequests } from "./assign/auto";
 import { createDevPoolPR } from "./pull-request";
 import { runOnPush, validateConfigChange } from "./push";
-import { findDuplicateOne } from "./issue";
+import { findDuplicateOne, embeddings } from "./issue";
 import { watchLabelChange } from "./label";
 
 export const processors: Record<string, Handler> = {
   [GithubEvent.ISSUES_OPENED]: {
     pre: [nullHandler],
     action: [findDuplicateOne, issueCreatedCallback],
-    post: [nullHandler],
+    post: [embeddings],
   },
   [GithubEvent.ISSUES_REOPENED]: {
     pre: [nullHandler],
@@ -44,12 +44,12 @@ export const processors: Record<string, Handler> = {
   [GithubEvent.ISSUE_COMMENT_CREATED]: {
     pre: [nullHandler],
     action: [handleComment],
-    post: [nullHandler],
+    post: [embeddings],
   },
   [GithubEvent.ISSUE_COMMENT_EDITED]: {
     pre: [nullHandler],
     action: [handleComment],
-    post: [nullHandler],
+    post: [embeddings],
   },
   [GithubEvent.ISSUES_CLOSED]: {
     pre: [nullHandler],

src/helpers/gpt.ts

@@ -178,3 +178,69 @@ export const askGPT = async (question: string, chatHistory: CreateChatCompletion
 
   return { answer, tokenUsage };
 };
+
+/**
+ * What is embedding?
+ *    An embedding is a vector of floating point numbers to measure the relatedness of text strings.
+ * How can I get an embedding using OpenAI?
+ *    To get an embedding, send your text string to the embeddings API endpoint along with a choice of embedding model ID (e.g., text-embedding-ada-002).
+ *    The response will contain an embedding, which you can extract, save, and use.
+ *
+ * Example Request:
+ *
+ *  curl https://api.openai.com/v1/embeddings \
+ *  -H "Content-Type: application/json" \
+ *  -H "Authorization: Bearer $OPENAI_API_KEY" \
+ *  -d '{
+ *     "input": "Your text string goes here",
+ *      "model": "text-embedding-ada-002"
+ *   }'
+ *
+ * Example Response:
+ *
+ * {
+ *  "data": [
+ *    {
+ *      "embedding": [
+ *        -0.006929283495992422,
+ *        -0.005336422007530928,
+ *        ...
+ *        -4.547132266452536e-05,
+ *        -0.024047505110502243
+ *      ],
+ *      "index": 0,
+ *      "object": "embedding"
+ *    }
+ *  ],
+ *  "model": "text-embedding-ada-002",
+ *  "object": "list",
+ *  "usage": {
+ *    "prompt_tokens": 5,
+ *    "total_tokens": 5
+ *  }
+ * }
+ * @param words - The input data to generate the embedding for
+ */
+export const generateEmbeddings = async (words: string): Promise<number[]> => {
+  const logger = getLogger();
+  const config = getBotConfig();
+
+  if (!config.ask.apiKey) {
+    logger.info(`No OpenAI API Key provided`);
+    throw new Error("You must configure the `openai-api-key` property in the bot configuration in order to use AI powered features.");
+  }
+
+  const openai = new OpenAI({
+    apiKey: config.ask.apiKey,
+  });
+
+  const embedding = await openai.embeddings.create({
+    // TODO: A couple of embedding models exist and `text-embedding-ada-002` is the one recommended by OpenAI
+    //  because it's better, cheaper and simpler to use.
+    // We might need to move the hardcoded model: `text-embedding-ada-002` to the bot configuration for better extensibility
+    model: "text-embedding-ada-002",
+    input: words,
+  });
+
+  return embedding.data[0]["embedding"] as number[];
+};

src/helpers/index.ts

@@ -10,3 +10,4 @@ export * from "./payout";
 export * from "./file";
 export * from "./similarity";
 export * from "./commit";
+export * from "./gpt";

supabase/migrations/20231030085814_create_embedding_table.sql

@@ -0,0 +1,11 @@
+-- Creates the `embeddings` table to store an embedding for every issue
+-- We've decided to use the `text-embedding-ada-002` model in the beginning which has 1536 output dimensions
+CREATE TABLE IF NOT EXISTS embeddings (
+    id serial PRIMARY KEY,
+    org character varying(255) NOT NULL,
+    repo character varying(255) NOT NULL,
+    issue integer NOT NULL,
+    embedding vector(1536) NOT NULL, 
+    created_at timestamptz NOT NULL,
+    updated_at timestamptz NOT NULL,
+);