Aditya Jami

New feature in LlamaExtract 🔥 - an Auto-Schema Generation Agent 🤖 A big friction point in automating document extraction with LLMs is defining the schema in the first place. Iterating on all the JSON fields can be painful. Now you can auto-generate an initial schema on the fly. Check it out! LlamaExtract: https://lnkd.in/gAPk3emJ

166

4 Comments

🚀 Big News: Observe.AI has acquired Dubdub.ai to accelerate our leadership in 𝗦𝗽𝗲𝗲𝗰𝗵 𝗮𝗻𝗱 𝗩𝗼𝗶𝗰𝗲 𝗔𝗜 𝘁𝗲𝗰𝗵𝗻𝗼𝗹𝗼𝗴𝗶𝗲𝘀! We believe the 𝗳𝘂𝘁𝘂𝗿𝗲 𝗼𝗳 𝗵𝘂𝗺𝗮𝗻-𝗺𝗮𝗰𝗵𝗶𝗻𝗲 𝗶𝗻𝘁𝗲𝗿𝗮𝗰𝘁𝗶𝗼𝗻 𝗶𝘀 𝘃𝗼𝗶𝗰𝗲, and we’re proud to be pioneering that revolution—starting with the 𝗖𝘂𝘀𝘁𝗼𝗺𝗲𝗿 𝗘𝘅𝗽𝗲𝗿𝗶𝗲𝗻𝗰𝗲 𝗶𝗻𝗱𝘂𝘀𝘁𝗿𝘆. From day one, Observe.AI has led with innovation in speech processing and understanding—from proprietary speech-to-text, to emotion and sentiment detection—powering Real-Time Agent Assist, Speech Analytics, and Auto QA for 350+ enterprise customers. With DubDub.ai, we're now doubling down on 𝗖𝗼𝗻𝘃𝗲𝗿𝘀𝗮𝘁𝗶𝗼𝗻𝗮𝗹 𝗔𝗜, bringing state-of-the-art Text-to-Speech capabilities that prioritize 𝗵𝘂𝗺𝗮𝗻𝗻𝗲𝘀𝘀, 𝗻𝗮𝘁𝘂𝗿𝗮𝗹𝗻𝗲𝘀𝘀, and 𝗽𝗲𝗿𝘀𝗼𝗻𝗮𝗹𝗶𝘇𝗮𝘁𝗶𝗼𝗻 in every interaction. This is key to building trust in VoiceAI—something sorely lacking in today’s consumer experiences with legacy conversational AI technologies. Welcoming Rahul Garg Anchal Jaiswal Rahul Sankhwar Anubhav Singh and rest of the Dubdub team. The future is voice. The future is human + AI. Read more at: https://lnkd.in/gYfbGaqv

643

40 Comments

Building AI agents is catnip for most engineers, and it has never been easier to whip up a quick demo. But building an agent that can reliably handle millions of customer interactions is hard. It's why so many companies — including those with deep engineering and AI expertise — have chosen to build with Sierra vs. roll their own. https://lnkd.in/g2TrgZsy

245

9 Comments

⚠️ Founders - consider the below reasons to not use Forward Deployed Engineers (FDEs) to acquire your *first* customers. This signals (1) lack of PMF; (2) miscalculation b/w development and sales timeframes; and (3) mischaracterization of Palantir Technologies timing when they deployed their FDE model. 💡 At the early-stage, it's crucial to demonstrate ability to repeatedly acquire customers with limited resources - capital, team, & time - that supports the narrative of 'more funding will amplify my traction.' Bespoke solutions for one company demonstrates zero repeatability. These custom solutions will not carry over to other customers within the same cohort. 💡 AI accelerates engineering but sales, system integration, and stakeholder buy-ins are perpetually slow. This imbalance results in concentrating all your engineering resources into one company, ignoring other opportunities, and still not going live on an accelerated timeline. 💡 Palantir didn't start using FDEs until year 3 when they got their first *gov't* customer, not their first customer. Gotham was already built and had strong PMF. Gotham allowed Palantir to take GTM risk by using FDEs b/c if it didn't work, Gotham was humming in the background. TLDR: At the early-stage, repeatable customer acquisition is the holy grail. Don't let AI deceive you that it can accelerate human behavior like it has with development cycles. It C.A.N.N.O.T.

32

1 Comment

In a recent podcast hosted by Dwarkesh Patel, Andrej Karpathy brings a refreshingly rational take to the AI agent hype, reminding us how far we still have to go. The biggest bottlenecks: agents aren’t yet truly intelligent, multimodal, or capable of remembering what you tell them. They’re cognitively lacking, and it’ll likely take a decade to fix that. LLMs are incredible at autocomplete, but still fall short of cognition: they forget context, their knowledge is “hazy”, and their short-term memory is limited. As Karpathy notes, we’ve barely replicated parts of the brain like the hippocampus (memory) or amygdala (instincts). Humans don’t think in tables or documents, we think in relationships: “Person X works at Company Y as Title Z.” That’s why we designed SurrealDB to better represent how humans think. SurrealQL lets developers model the world the way people actually think - connecting entities, context, and time through a unified query layer that blends graph, document, relational, and time-series data with search and retrieval. The result: a foundation for agent memory: persistent, contextual, and multimodal. 👉 If you’re building agents that need to think and remember, it’s time to build on SurrealDB. https://lnkd.in/ew2qetmG

133

Today, DeepSeek AI released a new open-source OCR model that’s been making waves on X. At Pulse, we tested it on real-world documents like financial statements, handwritten forms, and nested tables. The results were interesting but inconsistent. The same challenges we’ve seen with other LLM-driven OCR systems showed up again: unstable outputs, hallucinated text, and unreliable structure preservation that make enterprise-scale deployment difficult. Let’s put that into context. With 1,000 pages containing 200 extracted fields each, even a 99 % accuracy rate means 2,000 incorrect entries. When precision is critical, that’s the difference between a trusted data pipeline and a broken one. Our tests revealed: • Bounding boxes that shifted across runs • Handwriting misreads and hallucinated text • Table structures that failed to render correctly Pulse’s approach combines proprietary table-transformer models with classical vision algorithms, using LLMs only where they excel. The result is near-perfect accuracy, full data preservation, and low-latency processing at scale. Want to chat? Send a message directly or use the link in the comments.

301

25 Comments

Question to those using LLMs every day: Which LLM work best for team collaboration and shared AI workflows? I'm personally split 50/50 between ChatGPT and Claude. But as we look to invest in an Enterprise account for Remento, I'm realizing I definitely have some blind spots... Our primary goal is to easily create reusable AI assistants that any team member can access, trained on our specific knowledge base. For example: 👉 Marketing copy aligned with brand voice 👉 Product feedback based on customer research 👉 Content strategy from audience data For those who've made this transition: what's working?

25

13 Comments