HOTG.AI

Multiversal Ventures has been nothing short of a wild ride. Wow ... three years! What I’m most grateful for is that we’ve been able to bring an incredible team along for the journey. I’m constantly in awe of what this group can do — their ability to deliver, again and again, makes me proud beyond words. Today, we’re 20 people strong. It’s hard to believe that just three years ago, it was just Akshay Sharma and me, asking the same question we still ask after every exit, every meeting, and every moment of impact: “So, what did we learn?” Multiversal Ventures was born after a successful exit from HOTG.AI. That experience taught us to trust in our own abilities, to respect luck and timing, and to keep moving forward with curiosity. And even now, as we scale, experiment, and build, we’re still asking that same question. We are just beginning.

This Sunday (20th July) marks 3 years of Multiversal Ventures. We are super proud of what we have been doing at MVV:  • Brought back our core AI team from HOTG.AI, doc.ai , and have since then added some of the best talent from our team's network  • Found our missions and purpose - AI in regulated markets (healthcare, insurance, finance, lending, and more) - the real "multiverse"  • Changed the company parameter to: "What if we built a company that exists forever"? We are able to transform over and over now with mvv-agentic  • The incubation + spinout model to support growth - launching a few spinouts - Fenero AI, Orfeo, {more coming}  • Created an AI industrial lab - hammer.ai - its first product launching this year - Anvil - https://trustanvil.ai/  • Created "Gravity" for working on interesting and hard problems for us to go on forever We are very early in building a generational company. This would not be possible without the team we love working with at Multiversal Ventures - thank you! I want to share a quote from Richard Dawkins that keeps us going as builders: 𝘛𝘩𝘦 𝘮𝘰𝘴𝘵 𝘱𝘰𝘸𝘦𝘳𝘧𝘶𝘭 𝘥𝘳𝘪𝘷𝘦 𝘪𝘯 𝘵𝘩𝘦 𝘢𝘴𝘤𝘦𝘯𝘵 𝘰𝘧 𝘮𝘢𝘯 𝘪𝘴 𝘩𝘪𝘴 𝘱𝘭𝘦𝘢𝘴𝘶𝘳𝘦 𝘪𝘯 𝘩𝘪𝘴 𝘰𝘸𝘯 𝘴𝘬𝘪𝘭𝘭. 𝘏𝘦 𝘭𝘰𝘷𝘦𝘴 𝘵𝘰 𝘥𝘰 𝘸𝘩𝘢𝘵 𝘩𝘦 𝘥𝘰𝘦𝘴 𝘸𝘦𝘭𝘭 𝘢𝘯𝘥, 𝘩𝘢𝘷𝘪𝘯𝘨 𝘥𝘰𝘯𝘦 𝘪𝘵 𝘸𝘦𝘭𝘭, 𝘩𝘦 𝘭𝘰𝘷𝘦𝘴 𝘵𝘰 𝘥𝘰 𝘪𝘵 𝘣𝘦𝘵𝘵𝘦𝘳. 𝘠𝘰𝘶 𝘴𝘦𝘦 𝘪𝘵 𝘪𝘯 𝘩𝘪𝘴 𝘴𝘤𝘪𝘦𝘯𝘤𝘦. 𝘠𝘰𝘶 𝘴𝘦𝘦 𝘪𝘵 𝘪𝘯 𝘵𝘩𝘦 𝘮𝘢𝘨𝘯𝘪𝘧𝘪𝘤𝘦𝘯𝘤𝘦 𝘸𝘪𝘵𝘩 𝘸𝘩𝘪𝘤𝘩 𝘩𝘦 𝘤𝘢𝘳𝘷𝘦𝘴 𝘢𝘯𝘥 𝘣𝘶𝘪𝘭𝘥𝘴, 𝘵𝘩𝘦 𝘭𝘰𝘷𝘪𝘯𝘨 𝘤𝘢𝘳𝘦, 𝘵𝘩𝘦 𝘨𝘢𝘪𝘦𝘵𝘺, 𝘵𝘩𝘦 𝘦𝘧𝘧𝘳𝘰𝘯𝘵𝘦𝘳𝘺. 𝘛𝘩𝘦 𝘮𝘰𝘯𝘶𝘮𝘦𝘯𝘵𝘴 𝘢𝘳𝘦 𝘴𝘶𝘱𝘱𝘰𝘴𝘦𝘥 𝘵𝘰 𝘤𝘰𝘮𝘮𝘦𝘮𝘰𝘳𝘢𝘵𝘦 𝘬𝘪𝘯𝘨𝘴 𝘢𝘯𝘥 𝘳𝘦𝘭𝘪𝘨𝘪𝘰𝘯𝘴, 𝘩𝘦𝘳𝘰𝘦𝘴, 𝘥𝘰𝘨𝘮𝘢𝘴, 𝘣𝘶𝘵 𝘪𝘯 𝘵𝘩𝘦 𝘦𝘯𝘥 𝘵𝘩𝘦 𝘮𝘢𝘯 𝘵𝘩𝘦𝘺 𝘤𝘰𝘮𝘮𝘦𝘮𝘰𝘳𝘢𝘵𝘦 𝘪𝘴 𝘵𝘩𝘦 𝘣𝘶𝘪𝘭𝘥𝘦𝘳. - 𝘧𝘳𝘰𝘮 𝘵𝘩𝘦 𝘣𝘰𝘰𝘬 𝘛𝘩𝘦 𝘈𝘴𝘤𝘦𝘯𝘵 𝘖𝘧 𝘔𝘢𝘯 𝘣𝘺 𝘉𝘳𝘰𝘯𝘰𝘸𝘴𝘬𝘪, 𝘑𝘢𝘤𝘰𝘣. #build

𝘾𝙪𝙧𝙧𝙚𝙣𝙩 𝙡𝙚𝙣𝙙𝙞𝙣𝙜 𝘼𝙄 𝙩𝙤𝙤𝙡𝙨 𝙚𝙭𝙩𝙧𝙖𝙘𝙩 𝙙𝙖𝙩𝙖 𝙗𝙚𝙖𝙪𝙩𝙞𝙛𝙪𝙡𝙡𝙮 (𝙣𝙤𝙩!) Then you manually re-enter it into 3 other systems. Back and forth conversations on the data creates more auxiliary information that goes missing. That's not automation. That's expensive data entry with extra steps. Real loan intelligence requires fully context engineered data per loan application. Coming soon from Fenero AI - Agentic AI for lending with Multiversal Ventures's validation system (Anvil) built in! #Agentic #AI #LoanIntelligence #Integration #Workflow

Researchers at NVIDIA, in their paper Small Language Models (SLMs) are the Future of Agentic AI - https://lnkd.in/g6UhYK6F, support a belief we have long held: the future of agentic systems lies not in monolithic LLMs, but in compositions of domain-specialized SLMs. SLMs fine-tuned for scoped tasks: like intent parsing, tool calling, or structured output can outperform generalists in latency, alignment, and cost. The paper details how 60-70% of LLM queries in real-world agents (like MetaGPT, Cradle) can be offloaded to smaller, cheaper, and more controllable models . Agent design is evolving - from central brains to distributed, specialized minds. Our previous work at HOTG.AI briefly covered this topic - https://lnkd.in/gsxYq9P7. We developed a portable intelligent container that enables smart orchestration. Now, we have integrated these ideas into Multiversal Ventures industrial research lab, hammer.ai #agentic #ai

At Multiversal Ventures's products (like Fenero AI) and services, we have built 100+ integrations, processed millions of documents, talked to dozens of companies to bring AI solutions. The core realization is that work still requires some command and control. Full agentic systems still require humans to give the go-ahead and reconcile. Every company aspires to invest in a new, shiny dashboard to keep stakeholders informed about the status of operations. However, the continuity of autonomous workflows often breaks down at this point. Surprisingly, email has stood the test of time. Each of our agents acts as a coworker who operates through your email, providing long-term memory and context for in-depth, threaded follow-ups. Email has become the new AI user experience for command centers. cc Akshay Sharma Kartik Thakore if you are interested in bringing our agentic systems alive for your company #agentic #AI

As a founder, a key focus for me is being an effective capital allocator. This involves investing in growth and monitoring the return on invested capital (ROIC). Now, with the emergence of agentic AI, we have a third aspect - opportunity to expand our reinvestment surface area without increasing operational expenses. How? By rethinking the traditional ways our company operates. Kartik Thakore and I have decided to transform Multiversal Ventures into a fully agentic organization. In his first piece, Kartik Thakore outlines our path to achieving zero-dollar operations. We're also expanding other areas where we aim for zero-dollar-x through the use of agents. Not having an exploding ops burden allows us to get maximum fungibility! Agentic AI companies benefit from a slipstream, enabling them to target multiple domains without reconstructing themselves each time. #agents #ai #agentic

We're building the first "agentic" operating system that runs companies so humans don't have to. Most companies scale by adding managers and meetings. We're doing the opposite—letting go and using AI agents to handle coordination automatically. Last week: 30 hours saved across our 8-person team. Our agent reads Git repos, shares updates, and eliminates status meetings without waiting for instructions. The hardest lesson? Learning to trust our emergent, self-adaptive agentic operating system. When you stop micromanaging every coordination point, something beautiful happens. My hands are healing. Our team builds instead of coordinates. We become human again. This is just the start of proving that letting go creates more control than holding tight ever could. Multiversal Ventures #Agentic #AI #FutureOfWork #ZeroOps #Automation #Leadership #Startups #Productivity

𝗔𝗰𝘁𝗶𝗼𝗻𝗮𝗯𝗹𝗲 𝗜𝗻𝘀𝗶𝗴𝗵𝘁𝘀, 𝗡𝗼𝘁 𝗝𝘂𝘀𝘁 𝗥𝗲𝗽𝗼𝗿𝘁𝘀 Validation is only valuable if it drives improvement. That's why our platform doesn't just identify issues - it provides actionable insights. ✨ When your agent 𝙢𝙞𝙨𝙝𝙖𝙣𝙙𝙡𝙚𝙨 𝙖 𝙨𝙘𝙚𝙣𝙖𝙧𝙞𝙤, you receive specific recommendations for improvement, not just an error report. These insights help engineering teams make targeted fixes and business teams understand potential impact. 💡 The result? Faster iteration cycles, more robust agents, and clearer communication between technical and business stakeholders. 🚀 Drop us a note at hammer@multiversal.ventures if you want to join the waitlist for trying this. #AIInsights #ContinuousImprovement #AgentOptimization #Anvil 🎉

As more of your Agentic providers adopt reasoning models it is crucial to know the tradeoffs from a risk and continuity of business. #businesslossfunction #reasoning

𝗤𝘄𝗲𝗻𝟯 𝘃𝘀 𝗗𝗲𝗲𝗽𝗦𝗲𝗲𝗸-𝗥𝟭: 𝗨𝗻𝗱𝗲𝗿𝘀𝘁𝗮𝗻𝗱𝗶𝗻𝗴 𝗔𝗜 𝗥𝗲𝗮𝘀𝗼𝗻𝗶𝗻𝗴 𝗠𝗼𝗱𝗲𝗹𝘀 𝗳𝗼𝗿 𝗕𝘂𝘀𝗶𝗻𝗲𝘀𝘀 Remember the buzz around DeepSeek-R1 and its impressive AI reasoning? Now, Qwen3 has entered the arena, pushing the boundaries even further! For companies adopting AI agents, understanding the different approaches these models take to "thinking" is key to the quality of products and services you'll get. 𝗪𝗵𝗮𝘁'𝘀 𝘁𝗵𝗲 𝗱𝗶𝗳𝗳𝗲𝗿𝗲𝗻𝗰𝗲 𝗮𝗻𝗱 𝘄𝗵𝗮𝘁 𝘀𝗵𝗼𝘂𝗹𝗱 𝘆𝗼𝘂 𝗸𝗻𝗼𝘄? 🧠 𝘘𝘸𝘦𝘯3: 𝘛𝘩𝘦 "𝘊𝘰𝘯𝘵𝘳𝘰𝘭𝘭𝘢𝘣𝘭𝘦 𝘛𝘩𝘪𝘯𝘬𝘦𝘳" Qwen3 offers a "thinking budget" – essentially allowing an AI agent (built on it) to dynamically adjust its reasoning effort. - Pro: Agents could be better tuned by providers – deep thinking for complex tasks, quicker responses for simpler ones. This can mean more fine control on unit economics of the agent. - Con: If this "budget" isn't optimized by the agent provider for specific use cases, the agent might seem slow, or its reasoning might be too shallow, impacting output quality. 💡 𝘋𝘦𝘦𝘱𝘚𝘦𝘦𝘬-𝘙1: 𝘛𝘩𝘦 "𝘓𝘦𝘢𝘳𝘯𝘦𝘥 & 𝘋𝘪𝘴𝘵𝘪𝘭𝘭𝘦𝘥 𝘛𝘩𝘪𝘯𝘬𝘦𝘳" DeepSeek-R1 showcased how AI can learn complex reasoning through Reinforcement Learning, with these smarts then effectively "distilled" into smaller, efficient models. - Pro: This could lead to AI agents that are surprisingly powerful yet run faster and more cost-effectively, enhancing the value you get. - Con: "Learned" reasoning, especially when distilled, needs rigorous validation by agent providers. Without it, there's a risk of unexpected behavior or a loss of nuanced understanding in the final agent product. 𝗙𝗼𝗿 𝗖𝗼𝗺𝗽𝗮𝗻𝗶𝗲𝘀 𝗔𝗱𝗼𝗽𝘁𝗶𝗻𝗴 𝗔𝗜 𝗔𝗴𝗲𝗻𝘁𝘀, 𝗪𝗵𝗮𝘁 𝗠𝗮𝘁𝘁𝗲𝗿𝘀? When your AI agent provider/vendor talks about leveraging these cutting-edge models, consider asking: 1. How does the underlying model's reasoning approach ensure consistent accuracy and reliability for our specific industry and tasks? 2. How is the balance between reasoning depth, speed, and cost being managed to optimize the quality of the agent's output for us? 3. What validation processes are in place to ensure these advanced reasoning capabilities translate into dependable agent performance? The advancements are exciting, promising more capable AI agents. The key is ensuring these sophisticated "brains" are harnessed effectively to deliver real-world product quality and value for your business. 𝘕𝘰𝘸 𝘸𝘦 𝘸𝘢𝘪𝘵 𝘧𝘰𝘳 𝘋𝘦𝘦𝘱𝘴𝘦𝘦𝘬 𝘙2 𝘱𝘢𝘱𝘦𝘳 :). Link to Qwen3 paper: https://lnkd.in/gEqQ5Hkh Link to Deepseek R1 paper: https://lnkd.in/g8qBCSy6 #AI #AgenticAI #LLMs #Qwen3 #DeepSeekR1 #cost