Subscribe for updates

Let's explore a real-world example of debugging a RAG-type application. While updating our company knowledge base, I discovered that basic queries received correct answers, but follow-up questions lacked appropriate context — conversational memory had stopped working. Here's how I traced the bug step by step using a mix of traditional and AI-specific debugging techniques.

At Focused Labs, we're not only about tech — we're about making a real impact in our community. Through employee donation matching, charitable prizes, and a laptop purchase program that gives back, we've built a culture where giving is woven into everything we do. Since 2022, we've matched over $17,000 in employee donations and we're not stopping anytime soon.

Are you searching for an exceptional framework to build custom AI applications? LangChain's latest update using LangChain Expression Language (LCEL) offers streaming, async support, parallel execution, and more right out of the box. Here's how we upgraded our RAG application to use LCEL — step by step, from whiteboard to production.

The holiday season is the perfect time to dive into new technology — and this year, that means building applications with AI. There's tons of content about cool AI startups and crafting the perfect prompt, but what about actually building production apps? Let's get technical and dive into LangChain Expression Language (LCEL), a composable interface that gives you streaming, async support, and parallel execution right out of the box.

OpenAI has ventured into Retrieval Augmented Generation with its latest offering, Assistants. Think ChatGPT, but with the ability to ask it about custom information. While it doesn't yet lead the pack in creating AI-driven knowledge bases, being able to spin up a custom chatbot in a couple of hours with roughly 75% accuracy is incredible. Here's what developers should know about its capabilities and limitations.

Since the announcement of GPTs, I've been itching to create one focused on accessibility. But as I delved deeper, I realized that simply replicating existing accessibility checkers wouldn't be enough. I pivoted to building a comprehensive design assistant — one that analyzes screenshots and offers insights based on WCAG guidelines, Nielsen Norman heuristics, UX laws, and material design principles from Google and Apple.

Interested in building an AI-powered app using generative models? A common challenge is latency — users waiting up to 30 seconds before seeing AI suggestions. By leveraging OpenAI's streaming feature with Server Sent Events, we cut the wait for first content from 30 seconds down to half a second. Here are two practical tips for enhancing your AI app's user experience through effective streaming.

Stepping into the world of AI, I quickly grasped the enormous potential of domain-specific chatbots. My team and I crafted a smart AI chatbot tailored for the Focused Labs website — not just a simple chat interface, but a comprehensive “Knowledge Hub” that leverages natural-language models. Here are the top six insights I uncovered during the process, from orchestrating multiple AI models to thinking less like a programmer and more like a linguist.

Why Focused Labs love Lightning Talks and how your organization can integrate them into your events as novel team building opportunities.

In the world of technology, it's easy to be lured by the promise of a brand new system. But tales of cost overruns, schedule slippages, and features left undelivered are common regardless of industry. That's why organizations should be looking for innovative ways to maximize their existing systems — tapping into underutilized modules and reengineering key components so old systems can produce new results.

Digital modernization gets all the attention, promising new technologies and efficiencies. But all that glitters isn't gold, especially in technology. When budgets are limited and business operations can't be interrupted by a lengthy cut-over, maximizing your existing legacy systems to their full potential can be the true win-win your team has been looking for.

“What should we build next?” is the essential product management question — and one of the hardest. Here are 13 non-mainstream techniques for uncovering your next killer product or feature idea, from studying customer misbehavior and running JTBD-style interviews to leveraging extreme questions and opportunity backlogs.

No matter how long you've been in the digital transformation consulting business, getting dropped in the middle of an organization to take on a big project can be daunting. There are politics, personalities, and a lot of people to juggle. But there's also plenty of opportunity if you know where to look. Here are actionable tactics for navigating enterprise organizations and delivering the best outcomes.

AI leverages both new and existing UX patterns; learn which patterns to use when designing AI interfaces.

Chatbots like ChatGPT run on Large Language Models that use statistics to figure out what words to string together. Sometimes they're wrong — and they don't know it. Those convincing but incorrect answers are called hallucinations. So how do you deal with them? From data cleaning and prompt engineering to LLM settings and source citations, here are the practical levers you can pull to improve accuracy.

This article introduces the concept of prompt engineering — and more specifically, programmatic prompt engineering — and how it relates to building applications with Large Language Models. Small changes to an LLM prompt can have a big impact on accuracy and value. We'll show how techniques like persona matching and instructor-based prompting can dramatically improve results without relying on the end user to craft the perfect question.