feat(examples): add the Christmas generator AI workflow example/chall…

…enge

examples/nextjs-christmas-dinner-generator-challenge/.env.example

@@ -0,0 +1,2 @@
+POSTGRES_URL=your_neon_database_url_here
+OPENAI_API_KEY=your_openai_api_key_here

examples/nextjs-christmas-dinner-generator-challenge/.gitignore

@@ -0,0 +1,42 @@
+# See https://help.github.com/articles/ignoring-files/ for more about ignoring files.
+
+# dependencies
+/node_modules
+/.pnp
+.pnp.*
+.yarn/*
+!.yarn/patches
+!.yarn/plugins
+!.yarn/releases
+!.yarn/versions
+
+# testing
+/coverage
+
+# next.js
+/.next/
+/out/
+
+# production
+/build
+
+# misc
+.DS_Store
+*.pem
+
+# debug
+npm-debug.log*
+yarn-debug.log*
+yarn-error.log*
+.pnpm-debug.log*
+
+# env files (can opt-in for committing if needed)
+.env*
+!.env.example
+
+# vercel
+.vercel
+
+# typescript
+*.tsbuildinfo
+next-env.d.ts

examples/nextjs-christmas-dinner-generator-challenge/README.md

@@ -0,0 +1,314 @@
+# Inngest Weekend challenge: Christmas Dinner AI Workflow
+
+Feeling adventurous this weekend? ⚡️
+
+Let's build a Christmas dinner generator workflow using Inngest and Neon using the following Hugging Face datasets:
+
+- [`Thefoodprocessor/holiday`](https://huggingface.co/datasets/Thefoodprocessor/holiday) is a dataset of 74k meals paired with specific holidays (ex: “Christmas”)
+- [`Thefoodprocessor/wine_type`](https://huggingface.co/datasets/Thefoodprocessor/wine_type) is a dataset 74k meal <> wine pairing (we’re in France! 🇫🇷)
+
+> [!NOTE]
+> The above datasets have been chunked and translated to json files, available in the Getting started section.
+
+You will find below instructions to get Neon and Inngest up and running with the provided datasets to complete the challenge.
+
+Happy hacking! 🤖 🎄
+
+## **Prerequisites**
+
+To complete this challenge, you will need:
+
+- Node.js and npm installed
+- A valid OpenAI API key (for embeddings and prompts)
+- A Neon account ([you can create one for free](https://console.neon.tech/) in under 2 minutes)
+
+## Getting started
+
+First, checkout this repository:
+
+```bash
+git clone [email protected]:inngest/inngest-js.git
+```
+
+And run the following command:
+
+```bash
+cd examples/nextjs-christmas-dinner-generator-challenge
+npm install
+```
+
+You will notice the following:
+
+- a `data/` folder containing the datasets as JSON files
+- a `scripts/` folder containing an `index.ts`
+
+Let’s start by indexing our datasets into a Neon vector database.
+
+### 1. **Setup our Neon vector database**
+
+Go to the [Neon console](https://console.neon.tech/) and open the SQL Editor to copy the following:
+
+```sql
+CREATE EXTENSION IF NOT EXISTS vector;
+
+CREATE TABLE IF NOT EXISTS "holiday_recipes" (
+    "id" serial PRIMARY KEY NOT NULL,
+    "content" text NOT NULL,
+    "metadata" jsonb NOT NULL,
+    "embedding" vector(512),
+    "created_at" timestamp with time zone DEFAULT now(),
+    "updated_at" timestamp with time zone DEFAULT now()
+);
+
+CREATE TABLE IF NOT EXISTS "wine_pairings" (
+    "id" serial PRIMARY KEY NOT NULL,
+    "content" text NOT NULL,
+    "metadata" jsonb NOT NULL,
+    "embedding" vector(512),
+    "created_at" timestamp with time zone DEFAULT now(),
+    "updated_at" timestamp with time zone DEFAULT now()
+);
+
+CREATE INDEX IF NOT EXISTS "embedding_idx"
+ON "holiday_recipes"
+USING hnsw ("embedding" vector_cosine_ops);
+
+CREATE INDEX IF NOT EXISTS "embedding_idx"
+ON "wine_pairings"
+USING hnsw ("embedding" vector_cosine_ops);
+```
+
+### 2. **Index our datasets into our Neon vector database**
+
+First, copy the `.env.example` file to `.env.local`.
+
+Then, from your [Neon console](https://console.neon.tech/) project dashboard, grab your _DATABASE_URL_ and add it to your local `.env.local` file.
+
+Finally, do the same for your OpenAI API Key.
+
+You can now open the `scripts/index.ts` file.
+
+The interesting bit is the `processJsonFile()` function that takes each dataset’s JSON file and indexes it in the Neon vector database.
+
+You can now run the datasets indexing command by running:
+
+```bash
+npm run index
+```
+
+This script takes up to 5 minutes to index the 20k records.
+
+In the meantime, let’s set up an Inngest workflow for our application.
+
+### **2. Setup Inngest and create a simple workflow**
+
+Our project comes with Inngest already installed and ready to be used in Next.js.
+
+Let’s walk through the main files:
+
+**`src/inngest/events.ts`**
+
+This file defines the events used to trigger our workflow.
+
+The project comes with the event later used in our example to trigger the Christmas dinner generation workflow.
+
+Events are composed of:
+
+- a mandatory `name`
+- an optional object as `data`
+
+The events are created with [`zod`](https://zod.dev/) and passed to the Inngest client, enabling an E2E-typed experience.
+
+**`src/inngest/functions.ts`**
+
+This is where our workflow is defined, using `inngest.createFunction()`.
+
+[Inngest Functions](https://www.inngest.com/docs/features/inngest-functions?ref=ai-hack-night-paris-2024) are composed of multiple steps, enabling you to write workflows relying on independently retriable steps.
+
+Our first workflow is a simple example of multiple “hello worlds”.
+
+**`src/server.ts`**
+
+Inngest Functions are exposed to the Inngest Dev Server via a simple express server.
+
+Let’s give a try at our sample Inngest workflow by starting the following commands (in separate terminal windows or tabs):
+
+_Start the Next.js app:_
+
+```bash
+npm run dev
+```
+
+_Start the local Inngest Dev Server:_
+
+```bash
+npm run inngest
+```
+
+You can now open the Inngest Dev Server at [http://127.0.0.1:8288](http://127.0.0.1:8288/) and trigger the workflow from the “Functions” tab by clicking on the “Invoke” button using the following event:
+
+```json
+{
+  "data": {
+    "participantsCount": 2,
+    "preferences": ["chicken", "chocolate dessert"]
+  }
+}
+```
+
+![Screenshot 2024-11-20 at 14.33.46.png](./readme-assets/Screenshot_2024-11-20_at_14.33.46.png)
+
+By switching to the “Runs” tab, you will see the workflow run in real-time, with all the information on each step input/output and status:
+
+![Screenshot 2024-11-20 at 14.34.24.png](./readme-assets/Screenshot_2024-11-20_at_14.34.24.png)
+
+> [!NOTE]
+> The Inngest DevServer is useful to trigger and rerun your workflow, and inspecting steps.
+> ⚠️ All `console.log()` performed in an Inngest Workflow are printed in the Express server terminal.
+
+Our datasets indexing should be complete; let’s update our Inngest Workflow by adding a sample Neon vector query and OpenAI call step!
+
+### **3. Add a first Neon vector query and OpenAI call in our Inngest workflow**
+
+Let’s now use our Neon vector database with some OpenAI calls into our Inngest workflow.
+
+We use Neon's `similaritySearchWithScore()` query method to perform a vector similarity search based on a query to find some Christmas recipes that match our preferences.
+
+Then, we use the search results to perform an OpenAI call to generate a meal plan.
+
+Replace the `src/inngest/functions.ts` with the following:
+
+```tsx
+import { openai } from "inngest";
+import { inngest } from "./client";
+import { loadVectorStore } from "../lib/vectorStore";
+
+
+export const generateMeal = inngest.createFunction(
+  { id: "generate-meal" },
+  { event: "meal.generate" },
+  async ({ event, step }) => {
+    const { participantsCount, preferences } = event.data;
+
+    const holidayRecipesVectorStore = await loadVectorStore("holiday_recipes");
+    const winesPairingsVectorStore = await loadVectorStore("wine_pairings");
+
+    // Query relevant holiday recipes based on preferences using vector similarity
+    const relevantRecipes = await step.run("query-recipes", async () => {
+      const query = `Christmas dinner recipes ${preferences.join(", ")}`;
+      const results = await holidayRecipesVectorStore.similaritySearchWithScore(
+        query,
+        10
+      );
+
+      return results.map(([doc, score]) => ({
+        recipe: doc.pageContent,
+        title: doc.metadata.title,
+        similarity: score,
+      }));
+    });
+
+    // Generate meal plan using step.ai.infer()
+    const mealPlan = await step.ai.infer("Generate a Christmas dinner menu", {
+      model: openai({ model: "gpt-4" }),
+      body: {
+        messages: [
+          {
+            role: "system",
+            content:
+              "You are a professional chef and meal planner. Create a detailed Christmas dinner menu based on the provided recipes and requirements.",
+          },
+          {
+            role: "user",
+            content: `Create a Christmas dinner menu for ${participantsCount} people with these preferences: ${preferences.join(
+              ", "
+            )}.
+
+      Use these recipes as inspiration:
+      ${relevantRecipes.map((r) => r.recipe).join("\n\n")}
+
+      Include:
+      1. Appetizers
+      2. Main course
+      3. Side dishes
+      4. Dessert
+      5. Estimated preparation time
+      6. Special instructions
+
+      Format the response in a clear, organized way.`,
+          },
+        ],
+        temperature: 0.7,
+      },
+    });
+
+    // other steps...
+
+    return {
+      mealPlan
+    }
+);
+
+export const functions = [generateMeal];
+```
+
+> [!NOTE]
+> Take some time to familiarize yourself with the APIs and concepts used in the above example:
+>
+> - Neon's `holidayRecipesVectorStore.similaritySearchWithScore()` helps in finding matching meals
+> - Inngest’s `step.ai.wrap()` helps getting better observability over your LLM calls, without having to `console.log()` them.
+
+Navigate back to the Inngest Dev Server at [http://127.0.0.1:8288](http://127.0.0.1:8288/) and rerun the last workflow run from the “Runs” tab.
+
+The Inngest Dev Server now highlights the steps relying on AI by showcasing some important information such as LLM inputs and outputs:
+
+![Screenshot 2024-11-20 at 14.59.42.png](./readme-assets/Screenshot_2024-11-20_at_14.59.42.png)
+
+### You’re all set!
+
+Your local project is now well set up and ready for action! You should also know enough to complete the next example.
+
+Feel free to pursue your own project if you feel inspired or adventurous! 🤘
+
+## Challenge
+
+Now that you played with Neon vector queries and the Inngest workflow steps, let’s try to build the following complete Christmas dinner generator.
+
+By taking a `"meal.generate"` Inngest event containing the number of participants and some preferences (ex: `[“chicken", "chocolate dessert"]`); our workflow will generate a Christmas dinner menu matching the provided preferences, add some wines pairing (_”french touch”_) and generate the associated shopping list:
+
+![image.png](./readme-assets/image.png)
+
+### Complete our Christmas meal workflow
+
+Take inspiration from the two first steps implemented in the “Getting started” section to add the following:
+
+1. **Wine Pairing Enhancement**
+   - Secondary RAG process for wine recommendations
+   - Uses menu context to find relevant wines
+   - Generates personalized pairing explanations
+2. **Shopping list generation**
+   - Uses menu context and suggests wines to generate a shopping list
+
+**Useful documentation links**
+
+- [Working with Inngest Workflow and steps](https://www.inngest.com/docs/features/inngest-functions/steps-workflows?ref=ai-hack-night-paris-2024)
+- [Neon `pgvector` extension](https://neon.tech/docs/extensions/pgvector)
+
+### Some additional ideas
+
+Use the following suggestions to improve your project for the demo part!
+
+**Combine additional datasets from [Thefoodprocessor](https://huggingface.co/Thefoodprocessor) to improve the generation relevancy**
+
+The [https://huggingface.co/Thefoodprocessor](https://huggingface.co/Thefoodprocessor) project lists multiple other datasets, such as allergy and ingredient alternatives. You can download an additional `.parcel` file into the `data/` folder and run the `npm run generate-jsons` script to generate new JSON files.
+
+Then, you can run the `npm run index` script again and integrate the new Neon vector database into your Inngest workflow! 🚀
+
+**Add more parameters to the `"meal.generate"` Inngest event and leverage Inngest step parallelism**
+
+Adding more input parameters to the workflow could also improve the relevancy of our Christmas meal generator.
+
+For example:
+
+- each participant could be an object with advanced criteria (food, wine, etc)
+- then, the Inngest workflow would generate a menu per person ([leveraging step parallelism](https://www.inngest.com/docs/guides/step-parallelism?ref=ai-hack-night-paris-2024)) and merge the final results into a unique menu

examples/nextjs-christmas-dinner-generator-challenge/app/actions.ts

@@ -0,0 +1,15 @@
+"use server";
+import { inngest } from "@/inngest/client";
+
+export async function generateMeal(formData: FormData) {
+  const participantsCount = Number(formData.get("participantsCount"));
+  const preferences =
+    formData.get("preferences")?.toString().split("\n").filter(Boolean) || [];
+
+  const { ids } = await inngest.send({
+    name: "meal.generate",
+    data: { participantsCount, preferences },
+  });
+
+  return ids[0];
+}