October 2024 Summaries

AI Engineering is maturing in 2024, focusing on providing safe and reliable AI experiences at scale. Three key pillars are used to achieve this mission: LLM Evaluation, Guardrails, and better orchestration. LLM Evaluation assesses the quality and relevance of AI model responses and continuously evaluates them over time. Guardrails help manage and control AI behavior, ensuring responses align with ethical and functional standards. Orchestration infrastructure is crucial for handling the increasing complexity of AI workflows, improving reliability and cost efficiency.

Vercel is a platform for developers creating globally distributed applications using React and the serverless approach. However, many users face limitations linked to the nature of serverless, as Vercel Functions are meant to serve direct user interactions or API calls and should be completed quickly. This article covers four solutions to solving Vercel Function timeout issues: 1. Bump your Vercel Function max duration by updating the vercel.json file. 2. Leverage Edge Function's infinite duration by transforming long-running code into a streamed response. 3. Move your code into a Durable Function, which enables running code asynchronously for extended periods. 4. Upgrade to a Vercel Pro or Enterprise plan to increase the default max duration and access additional features.

Next.js is a popular framework for TypeScript developers that simplifies the development process by merging front-end and back-end tasks, but it poses challenges on serverless platforms due to the default 10-second duration limit for serverless functions. Developers often seek to extend these durations for tasks like integrations or AI workflows. To address these limitations, several solutions are proposed: increasing the Vercel Function's max duration to 60 seconds through the vercel.json file, utilizing Vercel's Fluid Compute for pooling resources, adopting Durable Functions for asynchronous execution, or upgrading to Vercel's Pro or Enterprise plans for extended timeouts. Each of these approaches offers a way to manage long-running processes more efficiently while maintaining the benefits of serverless architecture.

Traditional queuing systems are falling short in managing complex workflows, lack built-in traffic control mechanisms, have limited support for multi-tenancy, lack robust job management and recovery capabilities, and often lack robust built-in observability tools. Inngest simplifies complex workflows by consolidating functions and steps into a single, intuitive system, offers native flow control, supports multi-tenancy eliminating the need for separate queues and complex permission setups, enhances job management with greater control and visibility, and provides comprehensive observability & metrics dashboard.

This article discusses a reimagined contacts importer leveraging the power of reasoning models with Inngest. It introduces an agentic workflow pattern that relies on model reasoning to achieve a goal instead of providing static prompts. The author presents a demo CRM Next.js application featuring an autopilot contacts import feature, demonstrating how agentic workflows can be used today to bring new capabilities to applications. The article also covers the challenges and unpredictability associated with adding agentic workflows to an application and provides solutions for managing these issues.

This article presents an example of using agentic workflows in a CRM contacts importer built with Next.js and OpenAI's o1 reasoning model. The author demonstrates how to design an autopilot contact importer feature by leveraging the new OpenAI o1 reasoning model to plan the best steps to import a given file, ensuring proper field settings, ranking provided contacts as good software Sales leads, and matching provided columns. The article also discusses the challenges of agentic workflows, such as duration volatility, reliability, and unpredictability, and how Inngest helps deal with these issues by bringing long-running and retriable workflows inside Next.js applications.

MEGA SEO is an AI-powered platform that automates blogging by handling everything from indexing websites to generating and publishing content. The complex multi-stage workflow involves tasks such as crawling, writing, editing, and publishing while maintaining high quality and relevance. Inngest, an event-based orchestration solution, was used to manage this process efficiently. It allowed for breaking down the complex process into manageable steps, running tasks in parallel, handling retries and errors, and triggering different parts of the workflow based on specific events. The platform also uses AI models for keyword research, article writing, image generation, and more.