Daniel C. Liem’s Post

"Fear of technical challenges holding you back? Here's how a non-technical marketer used AI to build a data enrichment webhook—without writing code. The key was combining accessible tools like Zapier with AI guidance. What technical hurdles could you overcome with the right AI support? Read our latest case study to learn how."

AI Future Is Already Here

Today's BuzzBelow post talks about Glean. Glean is an AI startup transforming corporate data accessibility. Their platform uses advanced technologies like RAG and integrates large language models for tailored knowledge graphs and chat assistants. Co-founded by Arvind Jain, T.R. Vishwanath, Piyush Prahladka, and Tony Gentilcore, Glean has raised over $200 million, offering easy implementation with over 100 connectors for businesses seeking efficient data analysis and decision-making. Read the post here: https://lnkd.in/gFqUA2YQ #GleanAI #Insights #StartupSuccess

Whether you're a CTO or a janitor, AI is reshaping everything we do, and now is the time to leverage its power strategically. Our CEO, Ilya Levtov, explores the rapid evolution of AI and its transformative impact on both enterprises and startups in his latest blog: https://lnkd.in/ghhqyuum From AI-powered agents to multi-modal workflows, the potential is immense, but the challenges are real. Learn how to stay ahead in this new AI-driven era.

Unlocking Generative AI's Potential: From Monolithic to Agentic The world is buzzing about Generative AI, but deploying it to production remains a hurdle for most enterprises. Limited inference infrastructure, data privacy concerns, security risks, and a lack of skilled personnel all contribute to this challenge. Currently, only 5-6% of businesses have successfully deployed generative AI use cases. The traditional monolithic model approach works for simple use cases, but struggles as complexity increases. This is where the "Agentic" approach offers a paradigm shift. Instead of monolithic models, Agentic systems build and deploy compound AI systems - systems that combine multiple components to solve intricate problems. Imagine this: A monolithic LLM is like a black box - input goes in, output comes out. An Agentic system is more like a team - it understands intent, gathers information, plans actions, and even utilizes tools like other LLMs, data stores, function calls, or existing APIs to complete a task. Agentic systems leverage your existing capabilities to deliver powerful results, but building them in-house requires significant expertise, time, and resources. This is where low-code/no-code Agentic frameworks come to the rescue. Several excellent low-code/no-code Agentic frameworks empower developers to bring generative AI to life: Google's Vertex AI agent Builder Microsoft's Autogen n8n- https://n8n.io/ Langchain - https://www.langflow.org/ CrewAI - https://www.crewai.com/ Relevance AI - https://relevanceai.com/ Memgpt These platforms provide intuitive user interfaces, making building Agentic systems surprisingly simple. These intuitive frameworks make building Agentic systems accessible to everyone. However, operationalizing these systems in production presents its own challenges. To address this, we need a repeatable and standardized process - AOPS (Agentic Ops) - to ensure efficient, consistent, and predictable deployment and support. AOPS enables us to fully unlock the power of Generative AI by making production deployment scalable, manageable, and secure. AOPS process is able to bring desperate teams together and single focus to build, fine-tune, deploy and operate Agentic system in production to solve complex use cases. #GenerativeAI #Agentic #AOPS #ProductionDeployment #LowCode #AIRevolution

Gemini 1.5: Empowering AI with Enhanced Performance Exciting news in the world of AI! Google DeepMind, led by CEO Demis Hassabis, is unveiling Gemini 1.5, the next generation of their groundbreaking Gemini model. Building upon the success of Gemini 1.0, this new version delivers significantly enhanced performance, revolutionizing AI data processing and understanding across modalities. Breakthrough Performance Enhancements: Gemini 1.5 represents a monumental leap forward in AI technology. With a new Mixture-of-Experts (MoE) architecture, this model is not only more efficient to train and serve but also offers a breakthrough in long-context understanding. These advancements pave the way for a more capable and efficient AI experience. Expanded Context Window: One of the most significant enhancements in Gemini 1.5 is its expanded context window. With a capacity of up to 1 million tokens, this model can process vast amounts of information in a single prompt. This capability opens up new possibilities for analyzing complex data sets, including 1 hour of video, 11 hours of audio, or codebases with over 30,000 lines of code. Improved Problem-Solving and Reasoning: Gemini 1.5 excels in problem-solving tasks across various modalities. It can analyze, classify, and summarize large amounts of content within a prompt, making it ideal for tasks such as code analysis and video understanding. Additionally, its enhanced reasoning capabilities allow it to extract meaningful insights from complex data sets. Enhanced Performance and Safety: In performance evaluations, Gemini 1.5 Pro outperforms its predecessor, Gemini 1.0 Pro, on 87% of benchmarks. It also maintains high performance levels even with its increased context window, demonstrating impressive "in-context learning" skills. Furthermore, Google DeepMind has conducted extensive ethics and safety testing to ensure the responsible deployment of Gemini 1.5. Availability and Future Plans: Gemini 1.5 Pro is currently available for early testing to a limited group of developers and enterprise customers via AI Studio and Vertex AI. While the model will initially be released with a standard 128,000 token context window, Google DeepMind plans to introduce pricing tiers that scale up to 1 million tokens in the future. For more details, visit: https://lnkd.in/gZUsjHK7 Follow me, Daniel Joseph Raj to learn more and stay updated! #ai #gemini15 #googledeepmind #aiinnovation #nextgenai #mixtureofexperts #longcontextunderstanding #airevolution

🚀 Why Multiple AI Calls Are Better Than One: A Case for Task Splitting 🤖 When tackling complex problems—whether you're working on code, analyzing text, or processing data—it's tempting to lump everything into a single AI request. But breaking tasks into smaller, focused steps often leads to better results. Let's explore why! 🔍 Better Clarity Through Task Splitting By dividing a large task into smaller subtasks, each can be approached with more precision. For example: Step 1: Analyze data or text to extract the main concepts. Step 2: Refine results by filtering irrelevant information. Step 3: Develop specific outputs, like summaries or solutions. This method allows the AI to focus on one task at a time, reducing errors and improving the quality of results. ⏳ Improved Efficiency and Focus Instead of delivering an "all-at-once" response, working incrementally allows each phase to build on the last. By breaking things down, you're more likely to spot issues early and fine-tune the process as you go, saving time and effort in the long run. 🛠️ How PlandexAI Leverages Task Splitting PlandexAI (https://plandex.ai/) specializes in breaking complex tasks into simpler steps, making the AI process more effective. When you give PlandexAI a coding problem or development challenge, it: Decomposes the task into smaller, manageable parts that are easier to handle. Assigns these subtasks to different processes, ensuring that each step is completed thoroughly before moving on to the next. This structured breakdown helps prevent confusion and inefficiencies, leading to more accurate and reliable results, whether you're building features, refactoring code, or analyzing data. By splitting tasks into simpler steps, PlandexAI maximizes the strengths of AI, enabling smoother, more efficient development! #AI #TaskSplitting #PlandexAI #CodeOptimization #Efficiency #ProcessImprovement #AgileDevelopment

I don't think I've spoken to a single company in the last few months that isn't claiming to be aggressively “adopting AI”. In an effort to stay on top of things, enterprises have shown a willingness to experiment with anything AI related, resulting in a tidal wave of pilots and POC’s across the entire AI stack. Unfortunately, the vast majority of these POC’s and pilots will never convert into enterprise contracts, which has become a gigantic distraction for AI founders today. This week, I've pulled together some of our learnings about how some of the most successful Enterprise AI companies in the world (e.g., Cohere, Covariant, Hebbia AI, Twelve Labs) have crossed the enterprise adoption chasm, and navigated the maze of experimentation / test and trial budgets: 1. Focus on a tight offering in the early innings 2. Leverage a tried and true Enterprise Framework (e.g., BANT) 3. Spend the time and energy to be a bit more consultative 4. Push hard for paid pilots / POC’s over unpaid ones, even if the dollars are small 5. Be maniacal about feedback loops 6. Always define what a successful pilot or POC conversion would look like beforehand 7. Leverage specific case studies as much as possible 8. For very, very large contracts, experiment with Forward Deployed Engineers 9. Don’t be afraid to fire customers 10. Revisit assumptions about your ICP and product offering at least once a month If you're interested, there's more detail on each of these learnings in the link below:

🌟 Exploring the Potential of Generative AI in Postman 🌟 Today, I'm excited to share our journey with Generative AI models and their transformative impact. AI has diverse applications, and we've been exploring its potential for some time. I believe generative AI will boost productivity, enhance software quality, and help us achieve our mission of empowering 100 million builders. Recently, we introduced AI-driven autocomplete for FQL, the Flows Query Language that facilitates building Flows. In our Postman workspaces, we utilize Postbot, our AI companion, to debug and understand APIs, accelerate test writing, and manage large datasets efficiently. Postbot significantly improves our API testing workflow by designing better test cases and leveraging its deep contextual understanding of APIs. With built-in suggestions, we can easily communicate our needs to Postbot, whether through quick prompts or detailed descriptions. Looking ahead, the potential of AI is immense. Soon, Postbot will be able to: 1. Write documentation for our API calls. 2. Build a test suite from scratch through the Collection Runner, streamlining the process of creating comprehensive test suites. 3. Summarize test data reports, enabling us to focus on the most critical insights. 4. Debug API calls on our behalf, ensuring they meet our intended outcomes. 5. Assist in searching for the right API calls within the context of our workflows. I'm thrilled about the future of AI and its invaluable role in our work. Stay tuned for more updates! #AI #GenerativeAI #Postman #APITesting #Productivity #Innovation

Last week, I set out to build a custom AI automation to streamline tasks for myself. In the process, I learnt a few insights that I think might be valuable to share with anyone looking to create their own AI solution. At first, I didn’t fully understand the difference between retrieval-augmented generation (RAG) and fine-tuning. Here’s a breakdown: With RAG, we essentially feed an AI model additional data that it can pull from as needed. Imagine you have a generative AI that answers questions about your business. By integrating a custom data source (like a database of company documents), the AI can “retrieve” the latest, most relevant information whenever a question comes up. This setup is cost-effective and allows your data to be automatically refreshed as it updates. On the other hand, fine-tuning takes an existing model (such as one from OpenAI) and customizes it using your own data, formatted in JSONL with specific input-output pairs that represent what you want the AI to learn. The goal is to make the AI more precise by training it to respond exactly how you’d like. However, fine-tuning can be costly because it requires retraining the model with your unique data. Some key takeaways for building your own AI tool: While RAG lets the AI use your data without deep customization, it may lack the depth and specific “know-how” that a fine-tuned model would have. If you need high accuracy tailored to very niche questions or tasks, fine-tuning might be the better option. But if you’re looking for broad functionality with fresh data integration, RAG can be a practical choice. Ultimately, it’s about balancing what you need with what your budget and timeline allow. Building this AI automation has been a great learning experience, which I will share soon in more detail! #AI #Automation #MachineLearning #GenerativeAI #AIDevelopment #TechInsights #FineTuning #RetrievalAugmentedGeneration #DataScience #AITools #BusinessAI #TechInnovation