Add databases use cases

devdays-2025-ai/.gitignore

@@ -173,6 +173,3 @@ dist
 
 # Finder (MacOS) folder config
 .DS_Store
-
-# Database
-local.db

devdays-2025-ai/README.md

@@ -54,3 +54,8 @@ bun run src/index.ts "country code"
   - Llama Index
   - Vercel AI SDK
 - [ ] Add more examples in db
+
+## Database
+
+You can connect to the database locally using an new SQLite Connection.
+Choosing the path of your local.db file for "Database path"

devdays-2025-ai/package.json

@@ -10,6 +10,7 @@
   },
   "dependencies": {
     "@libsql/client": "^0.14.0",
+    "@slack/web-api": "^7.8.0",
     "openai": "^4.73.1"
   }
 }

devdays-2025-ai/src/database/database.ts

@@ -0,0 +1,77 @@
+import { createClient, type Client } from "@libsql/client";
+import { embed } from "../openai";
+import data from "./data.json";
+
+export const database = createClient({
+  url: "file:local.db",
+});
+
+// In this sample codebase we rely on an libsql (sqlite with other stuff) database to store the errors and comments data and vectors that result from embeddings
+// If you want to explore vector datases, you can have a look at:
+// - Chroma: https://www.trychroma.com/
+// - Pinecone: https://www.pinecone.io/
+
+const sanitzedSQL = (message: string): string => {
+  return message.replaceAll("'", "");
+};
+
+export const reset = async (database: Client) => {
+  const dataWithEmbedding = await Promise.all(
+    data.map(async (item) => {
+      const threadsWithEmbeddings = await Promise.all(
+        item.threads.map(async (thread) => ({
+          ...thread,
+          embedding: await embed(thread.message),
+        }))
+      );
+
+      return {
+        ...item,
+        embedding: await embed(item.message),
+        threads: threadsWithEmbeddings,
+      };
+    })
+  );
+
+  await database.batch(
+    [
+      "DROP TABLE IF EXISTS comments",
+      "DROP TABLE IF EXISTS errors",
+      `CREATE TABLE IF NOT EXISTS errors (id INTEGER PRIMARY KEY AUTOINCREMENT, created_at DATETIME NOT NULL, title TEXT NOT NULL, message TEXT NOT NULL, embedding F32_BLOB(${process.env.EMBEDDING_DIMENSION}) NOT NULL)`,
+      `CREATE TABLE IF NOT EXISTS comments (id INTEGER PRIMARY KEY AUTOINCREMENT, created_at DATETIME NOT NULL, message TEXT NOT NULL, sender TEXT NOT NULL, embedding F32_BLOB(${process.env.EMBEDDING_DIMENSION}) NOT NULL, errors_id INTEGER NOT NULL REFERENCES errors(id))`,
+      "CREATE INDEX errors_idx ON errors (libsql_vector_idx(embedding))",
+      "CREATE INDEX comments_idx ON comments (libsql_vector_idx(embedding))",
+    ],
+    "write"
+  );
+
+  await database.execute("BEGIN TRANSACTION");
+
+  try {
+    for (const item of dataWithEmbedding) {
+      const result = await database.execute(
+        `INSERT INTO errors (created_at, title, message, embedding) VALUES ('${
+          item.created_at
+        }', '${sanitzedSQL(item.title)}', '${sanitzedSQL(
+          item.message
+        )}', vector32('[${item.embedding}]'))`
+      );
+      const errorId = result.lastInsertRowid;
+
+      for (const thread of item.threads) {
+        await database.execute(
+          `INSERT INTO comments (errors_id, created_at, message, sender, embedding) VALUES ('${errorId}', '${
+            thread.created_at
+          }', '${sanitzedSQL(thread.message)}', '${
+            thread.sender
+          }', vector32('[${thread.embedding}]'))`
+        );
+      }
+    }
+
+    await database.execute("COMMIT");
+  } catch (error) {
+    await database.execute("ROLLBACK");
+    throw error;
+  }
+};

devdays-2025-ai/src/database/slack.ts

@@ -0,0 +1,63 @@
+import { LogLevel, WebClient } from "@slack/web-api";
+
+type ErrorMessage = {
+  created_at: Date;
+  title: string;
+  message: string;
+  sender: string;
+  threads: Omit<ErrorMessage, "title" | "threads">[];
+};
+
+const ERROR_COLOR = "a30200";
+const client = new WebClient(process.env.SLACK_TOKEN, {
+  logLevel: LogLevel.ERROR,
+});
+
+//mtg-jobs
+const channelId = "C02FW8DGG2F";
+
+try {
+  const result = await client.conversations.history({
+    channel: channelId,
+    limit: 200,
+  });
+
+  const messages = result.messages?.filter((message) =>
+    message.attachments?.some((attachment) => attachment.color === ERROR_COLOR)
+  );
+
+  const data: ErrorMessage[] = [];
+  for (const message of messages ?? []) {
+    if (!message.ts) {
+      continue;
+    }
+    const replies = await client.conversations.replies({
+      channel: channelId,
+      ts: message.ts,
+    });
+
+    const userReplies =
+      replies.messages?.filter(
+        (reply: any) => reply.subtype !== "bot_message"
+      ) ?? [];
+
+    data.push({
+      created_at: new Date(parseFloat(message.ts) * 1000),
+      title:
+        (message.app_id === "A03DJ82GYP8" //LaliloBot
+          ? message.text
+          : message.attachments?.[0].title) ?? "",
+      message: message.attachments?.[0].text ?? "",
+      sender: message.username ?? message.user ?? "",
+      threads:
+        userReplies.map((reply) => ({
+          created_at: new Date(parseFloat(reply.ts ?? "0") * 1000),
+          message: reply.text ?? "",
+          sender: reply.user ?? "",
+        })) || [],
+    });
+  }
+  console.log(JSON.stringify(data));
+} catch (error) {
+  console.error(error);
+}

devdays-2025-ai/src/index.ts

@@ -1,4 +1,4 @@
-import { database } from "./database";
+import { database } from "./database/database";
 import { embed, generateMessage } from "./openai";
 
 const args = process.argv.slice(2);
@@ -11,7 +11,7 @@ const searchQuery = args.join(" ");
 const queryEmbed = await embed(searchQuery);
 // https://docs.turso.tech/features/ai-and-embeddings
 const distanceComputation = `vector_distance_cos(errors.embedding, vector32('[${queryEmbed.join(
-  ", ",
+  ", "
 )}]'))`;
 const result =
   await database.execute(`SELECT errors.id as id, errors.message as errorMessage, errors.created_at as createdAt, comments.message as comment, comments.created_at as commentCreatedAt, comments.sender as commentSender, ${distanceComputation} as distance
@@ -39,6 +39,6 @@ const data = Object.values(orderedGroups).map((group) => ({
 console.log(
   await generateMessage(
     "You are a specialized AI that reads Lalilo error reports and the comments written about them by the team. From a given JSON containing an error message and its comments, create a clear, brief summary sentence that captures the essence of the issue and its resolution. Focus on what happened and how it was resolved if that information is present. Also highlight any team member that could have context about it and the dates it happened.",
-    JSON.stringify(data, null, 2),
-  ),
+    JSON.stringify(data, null, 2)
+  )
 );