Aditya Sharma’s Post

I recently gave Langfuse a go — and it’s quickly becoming one of my favourite open-source tools for working with LLMs and AI agents. If you’re building complex AI systems, Langfuse helps you observe, debug, and improve how your models perform — from prompt tracing to evaluation and version control. What is Langfuse? Langfuse is an open-source LLM engineering platform for observability, debugging, evaluation, and performance tracking. It helps teams build, monitor, and iterate on LLM-based systems with full visibility into how each prompt, response, and tool call behaves. What can we do with it? • Trace and debug every model call, prompt, and chain step to understand where latency or errors occur. • Manage and version prompts directly, without redeploying your app. • Run evaluations on model outputs, compare results, and track accuracy over time. • Collect datasets and test cases to benchmark performance and catch regressions. • Experiment safely in a built-in playground before shipping new prompts or logic. Example use cases: 1. Debugging multi-agent workflows – see where decisions go wrong in complex chains. 2. Prompt A/B testing – evaluate cost, latency, and quality between prompt versions. 3. Monitoring for drift – detect when model behaviour changes unexpectedly. 4. Compliance and auditability – maintain a full trace of how decisions were generated. Why it matters? As AI systems get more complex, observability is no longer optional. Langfuse bridges the gap between prompt engineering and AI operations (AIOps) — helping teams move from experimentation to reliable, production-grade performance GitHub and resources: • Langfuse GitHub: https://lnkd.in/ed-pjRRP • Langfuse Examples: https://lnkd.in/ezFHBQ3k • Langfuse Docs: https://langfuse.com/doc #Langfuse #AI #LLM #AIEngineering #AIOps #AgenticAI #Observability #PromptEngineering #AIagents #OpenSource

-- Spring AI + MCP: Building Plug-and-Play AI Agents Excited to share my latest medium article where I dive into how Spring AI integrates with the Model Context Protocol (MCP) to enable modular, scalable, and intelligent AI agents. -- Learn how to: Use @McpTool, @McpPrompt, and @McpResource annotations Build MCP-compliant agents with Spring Boot Secure your AI workflows with best practices Whether you're exploring agentic AI or already building with Spring Boot, this post offers practical insights and code examples to get started. -- Read it here: https://lnkd.in/gutiu-TU -- Let me know what you think or how you're using Spring AI in your projects! #SpringAI #MCP #SpringBoot #AIAgents #Java #AgenticAI #PlugAndPlayAI #SoftwareArchitecture #OpenSource #MediumBlog

Big step toward democratizing access to knowledge and building the next generation of AI-powered applications. Perplexity is best AI alternative of Google search. It just provide better detailed search results , Its very different from other AI models which gives direct solutions with hallucinations and prevent humans to think deeply . If you want your critical thinking capabilities to growth then its better to use AI for real time accurate and detailed data only and calculate /decide conclusions yourself rather than complete solutions from a AI.

Your AI deserves more than limits. It's time to connect it to EVERYTHING. Storm MCP is an enterprise-grade MCP server gateway designed to seamlessly integrate your Large Language Model (LLM) applications with over 100+ integrations, powerful Retrieval-Augmented Generation (RAG) data sources, and tools. Why Storm MCP is the connection your AI needs: Seamless LLM + RAG Integration : Directly use the Storm Platform within Claude Desktop. Connect custom embedding models and vectorDB solutions for unparalleled RAG capabilities. Standardized Interaction Protocol : Provides one consistent and reliable spec to communicate with any data source, eliminating the need for custom integrations. Easy Tool Definition & Invocation : Define reusable tools like search, DB queries, and file reads , and call them directly from your model. Context Sharing & File Ops : Share session context across models and manage files directly within your AI workflows for more coherent and relevant responses. Open Source, Secure & Scalable: Built for high performance and scalability, with robust security and the flexibility of open-source code. Check out more: https://tryit.cc/Ws4Xc #AI #RAG #LLM #StormMCP #MLOps #Embeddings

There's a ton of great points in this article, which coincides with what we're seeing "in the streets"... 1. AI is not the answer to every problem. 2. You better get your data right, because if you don't you will just produce bad outcomes faster! 3. Governance, governance, governance - do it first, do it right, and ROI follows. #eisitconsulting #consulteis #datagovernance

I tested the Anthropic Claude Chrome Extension against an Observe, Inc. tenant with my go-to question: “Several users are reporting that payment is failing. Can you analyze logs, metrics, or traces in Observe from the past 24 hours to identify the cause?” Surprisingly, the results were just as good as what I’ve seen using Claude Desktop (Opus 4.1) + Observe MCP or AI SRE. We’re entering a stage where multiple assistants can support you during on-call investigations: AI SRE, your AI Agent (Claude Desktop, Cursor, Claude Code) with Observe MCP, the OpenAI Agent, and even the Claude Chrome Extension.

☝️ One more exciting thing from Airbyte 2.0: Meet Airbyte MCP In a world of AI agents, we're putting the power of data movement directly into your LLM workflows. Model Context Protocol (MCP) is becoming the standard for LLMs to leverage tools across applications. With Airbyte MCP servers, you can now: 🔧 𝗕𝘂𝗶𝗹𝗱 𝗰𝗼𝗻𝗻𝗲𝗰𝘁𝗼𝗿𝘀 without drowning in API docs 📊 𝗚𝗲𝗻𝗲𝗿𝗮𝘁𝗲 𝗽𝗶𝗽𝗲𝗹𝗶𝗻𝗲𝘀 directly from your AI interface 🔍 𝗗𝗶𝘀𝗰𝗼𝘃𝗲𝗿 𝗳𝗶𝗹𝗲𝘀 at sources automatically ⚙️ 𝗠𝗮𝗻𝗮𝗴𝗲 𝘄𝗼𝗿𝗸𝘀𝗽𝗮𝗰𝗲𝘀 through natural language ✅ 𝗔𝘂𝘁𝗵𝗼𝗿𝗶𝘇𝗲 𝗮𝗰𝘁𝗶𝗼𝗻𝘀 seamlessly The game-changer? Our connector builder MCP. Never feel a connector is out of reach again. Don't get trapped in documentation. Make your platform truly extensible, thanks to AI. As LLMs become more involved in coding, your data infrastructure needs to keep up. Airbyte MCP enables you to leverage all our functionality in the AI development experience you're already using. 🔗 https://airbyte.com/ai

Over the last few days, I’ve been learning and exploring LangChain — trying to understand how LLMs actually work and how we can interact with them through code. 🔍 Here’s what I did: 1️⃣ Patent Data Fetching – Used SerpAPI to fetch detailed patent data (example: Lithium Battery related patents). 2️⃣ Data Storage – Collected and stored raw JSON patent data locally. 3️⃣ Vector Database with FAISS – Converted patent abstracts into embeddings using HuggingFace Transformers and stored them in a FAISS vector DB for semantic search. 4️⃣ Query + Analysis – Enabled querying the database to find the most relevant patents and analyze them in a structured report. ⚡ Why this matters? Patent analysis is usually slow and manual. By combining AI + Vector Databases, I was able to build a system that can: Search patents semantically and embeddings (not just by keywords) Extract meaningful summaries (Title, Abstract, Publication Date, Patent ID) Provide clear insights in an easy-to-understand format 💡 This project was a great learning experience with LangChain, HuggingFace, FAISS, and SerpAPI. Excited to take this further into real-world applications like R&D assistance, legal tech, and innovation tracking. #AI #MachineLearning #LangChain #FAISS #VectorDatabase #PatentAnalysis #HuggingFace #Groq

𝗔𝗻𝘁 𝗚𝗿𝗼𝘂𝗽'𝘀 𝗟𝗶𝗻𝗴-𝟭𝗧 - 𝗢𝗽𝗲𝗻-𝗦𝗼𝘂𝗿𝗰𝗲 𝗧𝗿𝗶𝗹𝗹𝗶𝗼𝗻-𝗣𝗮𝗿𝗮𝗺𝗲𝘁𝗲𝗿 𝗠𝗼𝗱𝗲𝗹 The open-source AI race just accelerated. Ant Group released Ling-1T on October 9, 2025—a trillion-parameter LLM, fully open-sourced China's answer to GPT-5, and they're giving it away 𝗣𝗲𝗿𝗳𝗼𝗿𝗺𝗮𝗻𝗰𝗲: State-of-the-art on complex reasoning : • AIME 2025 Math: 70.42% accuracy (competitive with GPT-5) • Code generation: Matches GPT-5 • Logical reasoning: Balanced efficiency/accuracy 𝗧𝗵𝗲 𝗟𝗶𝗻𝗴 𝗳𝗮𝗺𝗶𝗹𝘆: Ant Group expanded their lineup : 1.Ling non-thinking models (including Ling-1T) 2.Ring thinking models (Ring-1T-preview—first open-source trillion-parameter thinking model) 3.Ming multimodal series 𝗪𝗵𝘆 𝘁𝗵𝗶𝘀 𝗺𝗮𝘁𝘁𝗲𝗿𝘀: 🌍 AGI as a public good Ant Group's CTO: "We believe AGI should be a public good—a shared milestone for humanity" 📊 dInfer framework Novel inference optimization for efficiency gains 𝗙𝗼𝗿 𝗱𝗲𝘃𝗲𝗹𝗼𝗽𝗲𝗿𝘀: Building on open-source infrastructure : • Full model weights available (no API dependency) • Self-hosting possible • Customization and fine-tuning enabled • No rate limits 𝗧𝗵𝗲 𝗴𝗲𝗼𝗽𝗼𝗹𝗶𝘁𝗶𝗰𝗮𝗹 𝗮𝗻𝗴𝗹𝗲: China is betting on open-source dominance : • Ant Group: Ling-1T • Alibaba: Qwen models • ByteDance, Baidu: Pushing open models Western labs remain largely proprietary. Open-source momentum is shifting East 𝗣𝗿𝗼𝗱𝘂𝗰𝘁𝗶𝗼𝗻 𝗶𝗺𝗽𝗹𝗶𝗰𝗮𝘁𝗶𝗼𝗻𝘀: Trillion-parameter models were proprietary until now : • Research access to frontier-scale architecture • Fine-tuning for domain-specific applications • Infrastructure learnings for scaling Source: Ant Group : https://lnkd.in/gafNpD2Y #OpenSource #LLM #TrillionParameter #AntGroup

In the future, software might not be based on code Today, software is rules-based. You press a button, and pre-coded functions determine what happens next. Anthropics "Imagine with Claude" reimagines this entirely. Instead of deterministic functions, Claude interprets what you're trying to do and generates the functionality (and UI) on demand. To put this in perspective, this is kind of what happened with self-driving cars. Early systems used thousands of rules: "IF RED LIGHT: BRAKE, ELSE: ACCELERATE" etc etc. But reality has infinite edge cases. So we switched to neural networks that learn principles instead of following rigid rules. Software might be next. What if instead of writing code to handle every possible action, AI just interpreted intent and generated the right response and updated the UI in real-time. This is what “Imagine with Claude” is trying to do, and the concept is really cool. When this gets faster and connects to MCPs, every piece of software becomes infinitely flexible. The line between user and developer blurs. This is a new era of software unfolding. https://claude.ai/imagine In the image, you’re seeing what Claude imagines to be a futuristic web browser. I used the web browser to go to “google.com”. Since “Imagine with Claude” is not connected to the internet, it imagined what google looks like. You can even use the imagined version of google to google cat photos.