Armagan Amcalar

I put together a step-by-step guide to help you implement LLM Optimisation the right way for E-commerce. In this carousel, you’ll find the high-level overview. The full version includes: → How to generate a structured product feed (CSV, JSON, XML) with the right attributes: title, GTIN, MPN, price, availability, and product URLs → How to apply Schema . org Product markup using JSON-LD, with full code examples → How to configure your robots.txt to allow GPTBot, OAI-SearchBot and PerplexityBot access to your PDPs and feeds → How to prerender JS-heavy product pages using static snapshots or SSR → Where and how to submit your feed to Perplexity’s Merchant Program and OpenAI’s early access initiative → How to debug AI visibility using server logs, schema validators, and live prompt testing inside LLM tools If you’re working on e-commerce, SEO, or growth, this is where things are moving. 💬 Comment or DM me, and I’ll send you the full guide. 📌 Save this post if you're not ready yet but know you'll need it.

580

914 Kommentare

One of the biggest misconceptions I hear about serverless is this: "It’s just a spaghetti of Lambda functions calling each other." Yes, I’ve seen that happen. When it does, it’s ugly, and it usually comes from inexperience and lack of design. But that’s not what serverless architectures are supposed to be. It’s like saying "cats are animals that poop on your bed", sure, accidents happen (…probably 😹), but that’s not the norm or the expected behaviour! So, what does a well-designed serverless architecture actually look like? From a bird’s-eye view, pretty much the same as what you'd build with containers or EC2: • Separate accounts per team/workload • System is decomposed into independent services • Every service owns its own data (no shared DBs) • Services are loosely coupled through events • Centralised logging and observability Whether I have an API (synchronous communication) or use events (asynchronous communication) does not depend on whether I use Lambda vs. containers. A serverless architecture doesn't have to be event-driven. Equally, an event-driven architecture can run on containers or EC2. Those are orthogonal architectural choices. Inside each service, I use the serverless-first mindset to decide on my tech stack, e.g. • Prefer DynamoDB over RDS • Prefer API Gateway of ALB • Prefer Lambda functions over containers or EC2 • Prefer EventBridge over Kafka The guiding principle is simple: pick the service that does the most heavy lifting. And with serverless technologies like Lambda, you get built-in multi-AZ redundancy; scalability; reduced attack surface; no need to manage infrastructure; simplified deployment; and, pay-per-use pricing. So no, you don’t expose "a bunch of Lambdas" as your service boundary. That’s not the goal. That’s just a mistake.

375

14 Kommentare

After MCP, A2A, etc., we have another Agent protocol (fully open-source, explained with code) ACP (Agent Communication Protocol) is a standardized, RESTful interface for Agents to discover and coordinate with other Agents, regardless of their framework. Here's how it works: - Build your Agents and host them on ACP servers. - The ACP server will receive requests from the ACP Client and forward them to the Agent. - ACP Client itself can be an Agent to intelligently route requests to the Agents (just like MCP Client does to tools). The document below details a step-by-step walkthrough of using ACP to allow two agents (built in different frameworks) to work together. Tech stack: - CrewAI as the first Agent. - Hugging Face smolagents as the second Agent - Qwen 3 as the LLM. - ACP as the communication protocol between them. We’ll create a research summary generator, where: - Agent 1 drafts a general topic summary (built using CrewAI) - Agent 2 fact-checks & enhances it using web search (built using Smolagents). As depicted in the last slide, ACP has enabled both Agents to communicate via a standardized protocol, even though they were built using different frameworks. Find a link to the GitHub repo with the code in the comments! ____ Find me → Avi Chawla Every day, I share tutorials and insights on DS, ML, LLMs, and RAGs.

404

23 Kommentare

AI is removing the biggest blocker for non-technical builders: 𝘭𝘦𝘢𝘳𝘯𝘪𝘯𝘨 𝘴𝘺𝘯𝘵𝘢𝘹. Before, building automations felt like learning kung fu the hard way. Endless hours studying obscure formulas, JSON paths, bracket nesting. You needed to spend considerable time on the (hopefully clear and up-to-date) documentation. One mistake and it all broke. This discouraged many wannabe builders. Now you only need to: a) 𝘬𝘯𝘰𝘸 𝘸𝘩𝘦𝘳𝘦 𝘵𝘩𝘦 𝘪𝘯𝘧𝘰 𝘪𝘴, and b) 𝘧𝘦𝘦𝘥 𝘪𝘵 𝘵𝘰 𝘈𝘐. AI can process new info like Neo in 𝘛𝘩𝘦 𝘔𝘢𝘵𝘳𝘪𝘹 - thanks to in-context learning. And for you, this means: no more syntax fear. You just say what you want: “𝘊𝘢𝘭𝘤𝘶𝘭𝘢𝘵𝘦 𝘥𝘢𝘺𝘴 𝘣𝘦𝘵𝘸𝘦𝘦𝘯 𝘰𝘳𝘥𝘦𝘳 𝘢𝘯𝘥 𝘥𝘦𝘭𝘪𝘷𝘦𝘳𝘺” → Airtable writes DATETIME_DIFF({Delivery Date}, {Order Date}, 'days') “𝘍𝘪𝘹 𝘵𝘩𝘪𝘴 𝘍𝘓𝘖𝘞 𝘧𝘰𝘳𝘮𝘶𝘭𝘢 𝘪𝘵’𝘴 𝘳𝘦𝘵𝘶𝘳𝘯𝘪𝘯𝘨 𝘦𝘳𝘳𝘰𝘳𝘴” → AI finds the misplaced parenthesis and rewrites: IF({Revenue} > 0, {Revenue} / {Leads}, 0) "𝘈𝘭𝘴𝘰 𝘢𝘥𝘥 𝘢 𝘤𝘰𝘯𝘥𝘪𝘵𝘪𝘰𝘯 𝘵𝘰 𝘦𝘹𝘤𝘭𝘶𝘥𝘦 𝘧𝘳𝘦𝘦 𝘵𝘳𝘪𝘢𝘭𝘴” → AI adds: IF(AND({Revenue} > 0, {Plan} != 'Free Trial'), {Revenue} / {Leads}, 0) “𝘏𝘦𝘳𝘦’𝘴 𝘵𝘩𝘦 𝘈𝘗𝘐 𝘥𝘰𝘤𝘴, 𝘣𝘶𝘪𝘭𝘥 𝘮𝘦 𝘢 𝘗𝘖𝘚𝘛 𝘳𝘦𝘲𝘶𝘦𝘴𝘵 𝘵𝘰 𝘤𝘳𝘦𝘢𝘵𝘦 𝘢 𝘯𝘦𝘸 𝘤𝘰𝘯𝘵𝘢𝘤𝘵” → AI reads the JSON schema, outputs a fully formed curl request or a fetch() function with headers, auth, and payload ready to run. “𝘖𝘯𝘭𝘺 𝘪𝘯𝘤𝘭𝘶𝘥𝘦 𝘱𝘢𝘪𝘥 𝘪𝘯𝘷𝘰𝘪𝘤𝘦𝘴 𝘧𝘳𝘰𝘮 𝘭𝘢𝘴𝘵 𝘮𝘰𝘯𝘵𝘩” → Make locates bundle[1].line_items[].status, builds a filter using formatDate and lastMonth() functions, no digging required. AI handles the structure. You focus on the problem. The bottleneck isn’t “How do I write this?” It’s become: “What should I build?”

72

28 Kommentare