Ranjith Sivaraman

Israeli observability startup Coralogix has raised $115 million in a Series E round, becoming a unicorn with a valuation of over $1 billion. Coralogix offers a full-stack observability and security platform that helps companies understand and resolve complex issues in their data streams. The company has developed an AI agent, Olly, which uses a semantic layer over internal and external data to automate data observability and provide sophisticated insights through simple text prompts. The all-equity round was led by NewView Capital, with participation from the Canada Pension Plan Investment Board (CPP Investments | Investissements RPC) and NextEquity Partners, alongside existing investors. The funding will be used to expand its engineering base in India, with a planned $100M investment over five years, and to further develop its AI agent. Congratulations to co-founder and CEO Ariel Assaraf and the Coralogix team. TechCrunch: https://lnkd.in/dS3NZrrG #Observability #Data #AI #DevOps #Unicorn #SeriesE #Funding #VentureCapital #IsraelTech

1

We're excited to partner with eventnest.net, a stealth event listing platform for colleges and universities across India. Through this collaboration, our data pipeline platform 'varCHAR' will power their data infrastructure - enabling them to design and manage data & API workflows seamlessly to meet their data needs.

4

Context demands a single source of truth, an independent universal layer untethered from siloed agents, to avoid fragmentation and enable true scalability in AI-driven decisions. Why does this matter? In agentic systems, isolated context leads to "confidently wrong" automations, as silos multiply errors. See my previous post: https://lnkd.in/gU8nPTiN From this lens, GitData.AI emerges as precisely that universal context layer for AI agents. Built on Git's foundational principles, it transforms repositories into a centralized, versioned hub where code, datasets, models, workflows, and their relationships aren't just stored but actively interconnected as a living graph. Agents don't hoard context internally; instead, they pull from this independent system—tracing provenance via commit histories, resolving conflicts through merges, and committing back refinements to compound collective intelligence.

5

Proud to see our team at AIQube Technologies showcasing the Agentforce ITSM Mobile Experience. This demo highlights how IT service management can become simpler, faster, and accessible directly from mobile devices, enabling quick incident creation and faster resolutions. Great work by the team putting this together. 🚀

15

I was leading a LLM Customization session today when someone jumped up and asked a question! Real musings from today's session on LLM customization: Someone jumped up mid-session and said — "Fine-tuning is just a few clicks now. But no one talks about the data prep that goes behind the scenes. THAT's the real slog." 1000% agree. And it's the thing that catches enterprises off guard every single time. Here's the uncomfortable truth: Customizing an LLM has absolutely gotten easier. The tooling is better. The UIs are friendlier. But data prep? Still a slow, expensive, enterprise-grade headache. AND NO AI LABS TALK ABOUT IT! Data scientists already spend ~80% of their time just cleaning and preparing data — before a single training run starts. For LLMs, add to that: collecting labeled examples, removing noise and inconsistencies, formatting for the model's input schema, splitting train/val/test sets, and ensuring domain relevance and diversity. And that's before you hit the real risks of fine-tuning: - Safety guardrail degradation (researchers at Princeton, Stanford, Virginia Tech and IBM found fine-tuned models could be induced to assist with harmful activities) - Catastrophic forgetting of general capabilities - Model version lock-in - Ongoing retraining costs every time your data distribution shifts So the first question I always ask a team that wants to fine-tune: Have you actually maxed out prompting? 9 times out of 10 — the answer is no. Before you go near a training pipeline, exhaust these first: → Few-shot prompting — 3–5 in-context examples can dramatically shift output quality → A real system prompt — role, constraints, output format, tone. Not a two-liner. → Chain-of-thought instructions — especially for reasoning tasks → RAG — if the gap is knowledge, not behavior, this is almost always the right answer Fine-tuning earns its place when: ✓ The task is genuinely outside the model's training distribution ✓ You need consistent behavior changes at production scale ✓ You have clean, high-quality, labeled data actually ready to go ✓ Prompting has plateaued after serious, structured iteration The tooling will keep getting easier. Data prep won't magically disappear. The teams moving fastest aren't the ones with the most custom models. They're the ones who know when NOT to fine-tune. --- Have you seen this pattern in your org? Where did data prep surprise you most? #GenerativeAI #LLM #PromptEngineering #FineTuning #DataPrep #MLOps #EnterpriseAI #AWS #AmazonBedrock #ArtificialIntelligence #AIEngineering #MachineLearning

23

Styled Utterances: Coming soon in Parlant 2.2. (GitHub: https://lnkd.in/dqex3nBH - https://parlant.io). For heavy duty, open-source Conversational AI. 1. Easily teach your agent (with examples) your style and tone of voice. The agent will then draft up what it wants to say, then revise that to match your style -- which is what the user sees. 2. Multiple messages: your agent will break down long messages into natural, standalone chunks. This is out of the box behavior with this new styled composition mode.

18

1 Comment

These products are not experiments. They are the result of years of real-world deployments, customer feedback, and deep domain expertise, built on the globally accepted Frappe Framework and ecosystem. We are simplifying tech, and our focus is on achieving GO-LIVE in 2 weeks. 

14

Running a Salesforce Partner firm taught me one thing growth is not about doing more, 𝙞𝙩’𝙨 𝙖𝙗𝙤𝙪𝙩 𝙙𝙤𝙞𝙣𝙜 𝙨𝙢𝙖𝙧𝙩𝙚𝙧 💻 𝙏𝙝𝙚𝙨𝙚 5 𝙎𝙖𝙡𝙚𝙨𝙛𝙤𝙧𝙘𝙚 𝙛𝙚𝙖𝙩𝙪𝙧𝙚𝙨 𝙗𝙚𝙘𝙖𝙢𝙚 𝙤𝙪𝙧 𝙨𝙚𝙘𝙧𝙚𝙩 𝙬𝙚𝙖𝙥𝙤𝙣𝙨: ✔️Flows that run tasks automatically ✔️Validation Rules that keep data clean ✔️Dashboards that give instant clarity ✔️Integrations that connect everything ✔️Automation that scales without stress They didn’t just make our work easier they redefined how we operate. If you want to scale with consistency, start mastering these today. #salesforce #mvclouds #automation #digitaltransformation #businessgrowth #efficiency #salesforcepartner #salesforceconsulting

72

The Hard Truth About Building Your Own SaaS Authentication ! This is for all the SAAS entrepreneurs & tech builders , coding their own epics of authentication systems, profile management, multi tenancy, billing, subscription management, to micro tasks such as user deletion, prorated billing, plan cancellation, prorating plan to billing, authenticating via Google, LinkedIn , Microsoft etc. etc. Trust me when I say it- for an early stage app or a new software product (< 10 Million $ ARR and < 50,000 users), it is never worth your time. I have been to "the build it all by ourselves" road for quite a few years now, and have painstakingly felt the burn of issues like user redirect, forgot password, MFA?, SMS APIs, state management, billing APIs failing and what not. Not to mention the big taxing matters of security (say implementing hash like Argon2/ bcrypt) , dev time , auth flow, the big Maintenance Debt that comes every month, UX wooes and the list goes on. Sigh - feeling breathless by now ! And then I moved to the so called Auth Libraries. Some of them are : - Clerk.com (top of the list in my book , using it as I speak) - Kinde (have tried it - super simple to setup!) -PropelAuth (similar to Clerk - org choice frankly depending on the billing plan) - Auth0 by Okta (the legend which started it all !) - SuperTokens (got introduced to them first time by Parag Arora and what a smooth flow for users!) + few other players. Out of all these, if I have to pick the big two right now - they would be Clerk & Kinde. For the simple fact, that they go beyond the routine authentication and help you do the following : - Pre-built UI for Auth and providers ranging from Google, Microsoft, LinkedIn, Facebook , Metamask (for Web3) and SSO. - Threat-protection (against brutes and bots). - Session management (UI view for admins). - B2B Org & Multi-Tenancy (users, teams, orgs, RBACs, profile switch etc.) - Integrated Billing. - SAAS Feature Management (have done this on Clerk which does it smoothly) - Email / SMS templates and more. I feel that Clerk / Kinde / usual suspects above are not a shortcut detour but an excellent gateway to begin with the bare basics you need to build a SAAS product. You may chose to either scale with them only or mature towards your own custom code for everything, once you are hitting the unicorn scale. Earlier these 3rd parties used to only provide Authentication (Auth0 early days!),but now, all of them are maturing into a Full Stack Auth & User Management System for Multi-Tenancy and beyond. And I am sure, more is yet to come! The old adage "Don't reinvent the wheel" applies sharply here. This isn't an endorsement as much as it is a post-implementation post of pure delight. :) PS: Screenshots in the image are of Clerk.com Clerk.com Kinde Okta PropelAuth #authenticationtech #MFA #clerk #auth0 #kinde #multitenancy #developingSAASapps #usermanagement #billingtech #subscriptiontech #3rdpartyauthentication

12

ambaricloud: 108 recipes. 🥭 https://lnkd.in/gS6fuAtu Ravi Kiran Gurram Anil Koripelly Bhushan Lakhe Raja Purkayastha When this recipe e-book was first published on January 1, 2025, our goal was to give engineers, architects, and data practitioners a hands-on guide to Apache Iceberg. In the months since, the pace of change has been remarkable. Platforms such as Snowflake, Databricks, Dremio, RisingWave, ClickHouse, Cloudera, Lakekeeper, Trino Software Foundation, and Amazon Web Services (AWS) have delivered significant new integrations, and Apache Iceberg itself has reached version 1.10.0, introducing capabilities that reshape how practitioners interact with the format. To reflect this momentum, we are undertaking a ground-up rewrite. Our priorities include: ✅ Iceberg 1.10.0 alignment and deprecation cleanup – removing outdated recipes that no longer reflect best practices while upgrading all demos, examples, and exercises to incorporate the changes introduced in Iceberg 1.10.0. For example, Snowflake’s support for catalog-linked databases to external catalogs makes earlier object integration recipes obsolete. ✅ Ecosystem validation – testing and refining recipes across Databricks, Snowflake, Dremio, and AWS to ensure seamless interoperability with Iceberg 1.10.0. We expect this rewrite to take two to three months. Once complete, this edition will provide readers with a technically accurate, future-ready reference for Iceberg 1.10.0 and its surrounding ecosystems--Snowflake, Databricks, Dremio, RisingWave, ClickHouse, Cloudera, Lakekeeper, Trino Software Foundation, and Amazon Web Services (AWS) Please 🕺 suggest 🔪 comment 🥃 recommend ✅ Spark Variant ---Iceberg1.10.0 spports now. https://lnkd.in/gBmaNrPH

11

4 Comments