User Experience

Don't just blindly use LLMs, evaluate them to see if they fit into your criteria. Not all LLMs are created equal. Here’s how to measure whether they’re right for your use case👇 Evaluating LLMs is critical to assess their performance, reliability, and suitability for specific tasks. Without evaluation, it would be impossible to determine whether a model generates coherent, relevant, or factually correct outputs, particularly in applications like translation, summarization, or question-answering. Evaluation ensures models align with human expectations, avoid biases, and improve iteratively. Different metrics cater to distinct aspects of model performance:  Perplexity quantifies how well a model predicts a sequence (lower scores indicate better familiarity with the data), making it useful for gauging fluency. ROUGE-1 measures unigram (single-word) overlap between model outputs and references, ideal for tasks like summarization where content overlap matters. BLEU focuses on n-gram precision (e.g., exact phrase matches), commonly used in machine translation to assess accuracy. METEOR extends this by incorporating synonyms, paraphrases, and stemming, offering a more flexible semantic evaluation. Exact Match (EM) is the strictest metric, requiring verbatim alignment with the reference, often used in closed-domain tasks like factual QA where precision is paramount. Each metric reflects a trade-off: EM prioritizes literal correctness, while ROUGE and BLEU balance precision with recall. METEOR and Perplexity accommodate linguistic diversity, rewarding semantic coherence over exact replication. Choosing the right metric depends on the task—e.g., EM for factual accuracy in trivia, ROUGE for summarization breadth, and Perplexity for generative fluency. Collectively, these metrics provide a multifaceted view of LLM capabilities, enabling developers to refine models, mitigate errors, and align outputs with user needs. The table’s examples, such as EM scoring 0 for paraphrased answers, highlight how minor phrasing changes impact scores, underscoring the importance of context-aware metric selection. Know more about how to evaluate LLMs: https://lnkd.in/gfPBxrWc Here is my complete in-depth guide on evaluating LLMs: https://lnkd.in/gjWt9jRu Follow me on my YouTube channel so you don't miss any AI topic: https://lnkd.in/gMCpfMKh

1 + 526 = 527. That’s not a math glitch. That’s the number of training rooms I’ve stepped into — and Greenlam Industries ltd proudly became the 527th. In an industry where finishes are polished, it’s time conversations are too. India’s building material and interior sector is on fire — growing rapidly with design-savvy clients who know what they want. And guess what they value just as much as the product? How you talk about it. In our latest session with Greenlam, we worked with cross-functional teams : sales, operations, and client-facing teams — all looking to master the art of impactful communication. Here’s what we unpacked together: The Minto Pyramid Principle How do you structure your message when you only have 60 seconds? This framework teaches you to lead with the conclusion and support it with clear, logical reasoning — top-down, like a pyramid. Crisp, confident, clear. Contrasting A powerful way to clarify your intent. “I’m not saying we ignore the client’s concern. I’m saying we address it after we finish the walkthrough.” One line. Instant clarity. No confusion. Conversational Landmarks These are guideposts that help structure your talk. Phrases like “Let me break that down,” “Here’s the big idea,” “Let’s wrap this up” — they anchor the listener, making your conversation easier to follow (and remember). Articulation Techniques We practiced slowing down, pausing smartly, and replacing jargon with relatable language. Because when you speak with precision, people listen with attention. These smiles? They came from a room full of professionals who walked in to learn communication — and walked out sounding more confident than ever. If you’re planning a program for your team across retail, real estate, interiors, or B2B let’s talk.

Founders who actually use their own product and become part of their target audience get to really understand the pains. Being a founder who uses your own product puts you in your customers' shoes. You see firsthand what works, what doesn’t and where the pain points are. This insider view is priceless because you really understand the needs and frustrations of your audience. When you live your users experience, you build real empathy. You feel their struggles and can create solutions that truly help. This goes beyond just data and survey — it’s about living the same experience. Using your product often helps you spot small but important fixes that might get missed otherwise. These little tweaks can really boost user satisfaction and product quality. Plus being an active user lets you connect with your community better. You join conversations and get direct feedback, keeping you in touch with your users' changing needs. So scratch your own itch and solve problems that you’ve personally experience because this can be a huge competitive advantage.

Case studies are great, but I often learn more from the dumpster fires. I don't think I'm alone. Last week's CMO Coffee Talk featured a variety of rebrand experience shares, and the vast majority of the most valuable lessons and takeaways came from mistakes. Even when you see, read or hear case studies presented, some of the most common questions are: ✔️ What would you do differently next time? ✔️ What do you wish you had known before starting the project or process? ✔️ What went wrong and what did you learn from that, and/or how did you pivot because of it? These all focus on lessons burn of failures. James Clear, author of Atomic Habits, says as much: "Stories of failure resonate more than stories of success. Few people reach the top, but everyone has failed—including those who eventually succeed. If you're teaching people how to succeed in a given field (or talking about your own success), start with how you failed." Most companies have case studies prominently featured on their Web sites and sales materials. What if you also included customer failures? We tried this once in a webinar series and it worked spectacularly. It exclusively targeted stalled opportunities - prospects who for some reason or another just weren't moving forward. We called the series "Customers Unplugged" or something like that. And in a live Q&A format we asked HARD questions. Things like: 💣 What do you regret about buying this product? 💣 What do you need new customers to know before they commit? 💣 What were some of the reasons you almost didn't buy? The exec team was terrified when we first proposed this. And yet, after each one we did, at least 3-4 large deals suddenly got unstuck. The world is not full of purely success stories. No prospect is going to believe your case studies represent 100 percent of your customer base. Be vulnerable to earn loyalty. Let more people hear your dumpster fires! I guarantee it will attract far more than it will repel.

Evaluations are extremely important for any AI application. That is, how do you know which models to use, if things are working optimally, etc? Today, we’re sharing a bit about our eval stack.   Behind every Decagon AI agent is a rigorous model evaluation engine built for the highest-stakes customer interactions.   When your agents are handling complex, customer-facing use cases, you need more than just promising model outputs. You need a framework that continuously and precisely measures real performance at scale.   In our latest blog post, we break down the core components of that evaluation framework: 🧠 LLM-as-judge evaluation – scoring real-world interactions across relevance, correctness, empathy, and naturalness, with human validation to catch edge cases 📊 Ground truth benchmarking – using curated, expert-labeled datasets to measure factuality and intent coverage 🚦 Live A/B testing – deploying variants in production and measuring their impact on real business outcomes like CSAT and resolution rate   This evaluation doesn’t stop once the latest version of an AI agent ships. Every insight feeds back into prompts, retrieval, and agent logic. The result: continuous improvement in the quality of customer experiences.   Check out the full blog in the comments.

Here is why leaderboards can fool you (and what to do instead) 👇 Benchmarks are macro averages, and your application is a micro reality. A model that’s top-3 on MMLU or GSM-Plus might still bomb when asked to summarize legal contracts, extract SKUs from receipts, or answer domain-specific FAQs. That’s because: 👉 Benchmarks skew toward academic tasks and short-form inputs. Most prod systems run multi-turn, tool-calling, or retrieval workflows the benchmark never sees. 👉 Scores are single-shot snapshots. They don’t cover latency, cost, or robustness to adversarial prompts. 👉 The “average of many tasks” hides mode failures. A 2-point gain in translation might mask a 20-point drop in structured JSON extraction. In short, public leaderboards tell you which model is good in general, not which model is good for you . 𝗕𝘂𝗶𝗹𝗱 𝗲𝘃𝗮𝗹𝘀 𝘁𝗵𝗮𝘁 𝗺𝗶𝗿𝗿𝗼𝗿 𝘆𝗼𝘂𝗿 𝘀𝘁𝗮𝗰𝗸 1️⃣ Trace the user journey. Map the critical steps (retrieve, route, generate, format). 2️⃣ Define success per step. Example metrics: → Retrieval → document relevance (binary). → Generation → faithfulness (factual / hallucinated). → Function calls → tool-choice accuracy (correct / incorrect). 3️⃣ Craft a golden dataset. 20-100 edge-case examples that stress real parameters (long docs, unicode, tricky entities). 4️⃣ Pick a cheap, categorical judge. “Correct/Incorrect” beats 1-5 scores for clarity and stability 5️⃣ Automate in CI/CD and prod. Gate PRs on offline evals; stream online evals for drift detection. 6️⃣ Iterate relentlessly. False negatives become new test rows; evaluator templates get tightened; costs drop as you fine-tune a smaller judge. When you evaluate the system, not just the model, you’ll know exactly which upgrade, prompt tweak, or retrieval change pushes the real-world metric that matters: user success. How are you’re tailoring evals for your own LLM pipeline? Always up to swap notes on use-case-driven benchmarking Image Courtesy: Arize AI ---------- Share this with your network ♻️ Follow me (Aishwarya Srinivasan) for more AI insights and resources!

I talk to 9 new CXOs almost every week. 7 out of the 9 struggle with this 1 thing: "Connecting Emotionally with Their Audience" This happens because they frequently rely too much on statistics and facts, which makes their speech impersonal and dry. The next thing they know, their capacity to inspire and lead their groups, stakeholders, and clients goes down to 0! And it’s not like they haven’t worked on it. Infact, at least 5 of those CXOs have experimented with different approaches to enhance their public speaking abilities, including going to seminars, reading books, and even rehearsing in front of a mirror. But these attempts were more about technique than on the emotional connection, which eventually made them give up. Once we SWOT analysed it all, finding the right approach was easy for us. What did we do? The Empathy-Driven Communication Approach: → Storytelling: We created gripping stories to illustrate the most important points to make the information memorable and relatable. → Analysis of the Audience: We concentrated on learning about the needs, feelings, and viewpoints of the audience. → Training in Emotional Intelligence: We aimed to improve their capacity to identify, control, and relate to their own feelings as well as those of their audience. The result? → 3X the Influence → 2X the Engagement → Stronger Relationships Today, they have transitioned from being data-driven presenters to influential storytellers who can connect deeply with their audience. Interested in transforming your public speaking skills and becoming an influential leader? DM me “INFLUENCE” P.S. What do you find most challenging about connecting with your audience during a presentation?

Okay, here is a disclaimer about this Tuesday post: This is not a rant (but an invitation for better listening). Every introduction I make about myself, half of those intros turn out to be an experience. Me: “Hi, I’m Meena, CEO of Namaste Data.” The other human in the interaction starts with: ● “I’ve been doing yoga for 15 years.” ● “I visited India 20 years ago!” ● “Oh! My cousin’s roommate’s uncle is from Nepal or maybe India?” ● “Namaste, I’m a yoga teacher.” ● “Congrats on your site. I just came from my yoga session. Namaste” All of them are far from questions like ● “What kind of data do you work with?” ● “Tell me more about the AI Equity Project.” ● “How do you help nonprofits with ethical data?” ● "I like that name. How did you come up with it?" I smile and nod — and also sigh a little. It’s like if someone told you their org was called “Orange Earth” and you replied: ● “I had an orange yesterday.” ● “Costco oranges are great.” ● “I bulk order oranges for my juice cleanse.” What if — and hear me out — we just paused? What if we seek more curiosity here? I don’t find these questions or comments offensive or annoying. Someone is trying to reach out in the best way they think – and I respect that intention. But what if, instead of finding common ground  – in the first intro statement -  through assumptions, we found it through thoughtful, pause-filled questions?...where we truly listened. And to listen better: ● Pauses are okay ● Questions are okay ● Seeking time to process what we heard is okay Maybe the intention in all our communication is a connection, but the connection deepens when, instead of finding commonalities based on better listening, being present in the conversation better. Let’s make space for conversations that go deeper than first impressions.

“Breaking down the anatomy of a perfect 60-second product demo” – written in the same tone, rhythm, and structure as your earlier examples 👇 Most 60-second product demos feel like 6 minutes. Too long. Too slow. Too much fluff. Here’s the truth: 📌 People aren’t watching to understand. They’re watching to decide. That’s why your demo can’t just explain your product. It has to sell it—in less time than it takes to skip a YouTube ad. So let’s break down the anatomy of a high-converting 60-second demo 👇 🔥 0-3 seconds: Hook or lose Your intro should say, “This product solves your problem.” → “Here’s how we removed dark spots in 30 days using X” → “Real results with zero downtime. See it in action.” ⚠️ No logos. No branding intro. Straight to pain or promise. 🎯 4-20 seconds: Problem & Promise Show the actual problem. Use real skin, not stock footage. And show how your product is the bridge between pain → solution. → Problem statement → Solution in motion → What makes it different 🎥 21-45 seconds: Demo in motion Use jump cuts. Show steps, not stages. Show results, not just process. ✅ Application shots ✅ Texture close-ups ✅ Progress/Before-After clips Bonus: Add subtle on-screen captions to guide viewer attention. 🧠 46-60 seconds: Social Proof + CTA People trust people. Not packaging. → Clip of a customer review → Quick doctor quote → Visual result with timestamp → Clear CTA: “DM us” | “Try it today” | “Shop now” End with momentum, not a fade-out. ✨ That’s a winning demo structure. And it works especially well for vertical skincare content. If you’re a founder, creator, or brand sitting on raw skincare footage— → We can turn it into clean, high-retention product videos like this. DM me “DEMO” if you want us to show you how. Because in 2025—clarity converts. Fancy doesn’t. 🎯

Our App Was Crawling at Snail Speed… Until I Made This One Mistake 🚀 A few months ago, I checked our Lighthouse scores—30s. That’s like running an F1 race on a bicycle. 🏎️➡️🚲 𝐀𝐧𝐝 𝐭𝐡𝐞 𝐰𝐨𝐫𝐬𝐭 𝐩𝐚𝐫𝐭? We did everything right—modern stack, top framework, best practices. Yet, our app was sluggish. ❌ AI-powered search engines ignored us. ❌ Users kept waiting. ❌ Something was off. So, we did what every dev does—optimize. 🔧 Cut dependencies 🔧 Shrunk bundles 🔧 Tweaked configs We went from 30s to 70s. Better, but still not great. Then, I made a 𝐦𝐢𝐬𝐭𝐚𝐤𝐞. A glorious, game-changing mistake. One deploy, I accidentally removed JavaScript. And guess what? Lighthouse: 91. 😳 Sure, nothing worked. No buttons, no interactivity. But it proved our app could be fast. 💡 The lesson? Stop making JavaScript do everything. 𝐒𝐨 𝐰𝐞 𝐫𝐞𝐛𝐮𝐢𝐥𝐭: ✅ JavaScript only where needed ✅ No unnecessary hydration ✅ No bloated client-side rendering 𝐓𝐡𝐞 𝐫𝐞𝐬𝐮𝐥𝐭? 🚀 From 30s to consistent 90+ scores 🚀 Faster load times 🚀 Better search engine visibility Sometimes, the problem isn’t a lack of optimization—it’s an excess of complexity. Not every app needs a heavy framework. Not every UI should be hydrated. If you’re struggling with performance, ask yourself: ❓ Do I really need this much JavaScript? ❓ Can I pre-render more? ❓ What happens if I strip everything back to basics? You might be surprised by what you find. 👀