Kevin Ngo

The 𝗣𝗮𝗹𝗮𝗻𝘁𝗶𝗿 "𝗠𝗮𝗳𝗶𝗮" is quietly building again — this time inside 𝗬 𝗖𝗼𝗺𝗯𝗶𝗻𝗮𝘁𝗼𝗿 Former Palantir engineers and operators are launching startups that touch AI infra, developer tools, and next-gen data intelligence. I pulled the data using Harmonic — my go-to startup discovery engine as a BA - and here are 6 new YC-backed startups founded by ex-Palantir talent 👇 ▫️ Clidey (YC X25) – AI-powered infra platform with containerized workflows and real-time analytics ▫️ Overstand Labs (YC W25) – AI insights from Slack, email, WhatsApp to surface product signals and churn risk ▫️ Laminar (YC S24) – LLM tracing + evaluation with blazing-fast Rust pipelines and graph-based workflows ▫️ Zeit AI (YC S24) – Natural language queries across messy enterprise data, with GDPR-safe profiling ▫️ Kenley– AI copilots for consulting workflows like RFPs, due diligence, and internal search ▫️ Parsewise – AI agents that extract structured data from messy documents, with traceability All 6 were discovered through Y Combinator’s latest batch. All have roots at Palantir. And all are solving real data problems at the edge of what's possible. Which one is your favourite? 

329

19 Comments

Just getting started with LLMs? Don’t let the jargon slow you down. This 5-part LLM Glossary is your shortcut to mastering the fundamentals👇 1️⃣ Model Types • Foundation Model: Pretrained on large-scale, diverse datasets to learn general capabilities. • Instruction-Tuned Model: Further trained to follow user instructions precisely. • Multi-modal Model: Handles multiple input/output formats like text, images, and audio. • Reasoning Model: Optimized for logic, problem-solving, and step-by-step thinking. • Small Language Model (SLM): Lightweight models for fast, efficient, on-device tasks. 2️⃣ Training & Fine-Tuning • Pretraining: Initial learning phase using massive, diverse data sources. • RLHF (Reinforcement Learning with Human Feedback): Aligns model behavior with human preferences. • DPO (Direct Preference Optimization): Uses ranked preferences instead of rewards for training. • Synthetic Data: AI-generated data used to supplement real datasets. • Fine-Tuning: Targeted retraining on specific domains or tasks. • LoRA / QLoRA: Techniques for efficient low-resource fine-tuning. • Guardrails: Rules or filters applied to enforce safety, ethics, and compliance. 3️⃣ Prompt Engineering • System/User Prompts: Define model behavior and user context in a session. • Chain-of-Thought (CoT): Prompts that guide the model to reason step-by-step. • Few-Shot / Zero-Shot Learning: Demonstrating tasks with few or no examples. • Prompt Tuning: Optimizing prompts via training for specific outcomes. • Context Window: The total number of tokens the model can "remember" per interaction. 4️⃣ Inference & Generation • Temperature: Adjusts output randomness; lower values are more deterministic. • Max Tokens: Limits the number of tokens in a generated response. • Seed: Controls reproducibility of outputs. • Latency: Time taken by the model to respond. • Hallucination: When the model generates factually incorrect but plausible-sounding information. 5️⃣ Retrieval-Augmented Generation (RAG) • Retrieval: Fetching relevant data from external sources before generation. • Semantic Search: Search based on meaning, not just keywords. • Embeddings: Vector representations of text used for similarity matching. • Chunks: Dividing documents into smaller segments for better retrieval. • Vector Databases (VectorDBs): Stores embeddings for fast and accurate search. • Reranking: Rescoring retrieved documents to prioritize relevance. • Indexing: Structuring data for efficient retrieval during generation. 🌟 Follow the AIKaDoctor (Free AI & Data Science Resources) channel on WhatsApp: https://lnkd.in/dCTCEKKc 📌Follow Habib Shaikh For more such content.

232

36 Comments

This week, Anthropic launched Claude for Life Sciences, its first real push to make AI part of how science gets done. Why this matters 👇 - Scientists spend hours each week buried in docs and data. Claude aims to give them that time back by summarizing literature, drafting protocols, and analyzing data. And with prebuilt connectors to Benchling, BioRender, and PubMed, it fits well into existing workflows. - Some researchers have been eager to use AI but hesitant to apply it without clear benchmarks. Claude’s most capable model, Sonnet 4.5, shows strong performance in lab protocol and bioinformatics tasks, making AI adoption in the field more credible and comfortable. - Anthropic also introduced Agent Skills, instructions that dictate how Claude tackles specific tasks. The first skill they’ve built performs single-cell RNA sequencing quality control using scverse best practices, a sign they’re prioritizing technical rigor. It's a strong step forward, but the rollout points to where pharma still is on the AI adoption curve. Instead of the self serve model offered to consumers, Anthropic is partnering with consulting firms like Deloitte, Accenture and KPMG to bring Claude to enterprise pharma. I can't help but wonder, if the goal is to meet scientists where they are, why would it take a consulting team to get started? 🤔 WDYT? Image Credit: Anthropic Agent Skills, announcement link in comments

85

3 Comments

Slowing down junior hiring because of AI is a major mistake. Why? Because we fundamentally underestimate who today’s new hires are and what they bring to the workplace. Recently, Adrian Masson and I interviewed junior new hires at Moderna, across multiple functions including R&D, manufacturing, and quality control. Their AI usage was far above the company average, and what they do with it left us humbled and amazed. Here's a glimpse: 🎙️ Their workday starts in voice mode, using AI to clear their minds, set priorities, and map out their time. They delegate tasks to agents, launch all sorts of automations, and constantly check the outputs, adjusting objectives as new data comes in. 🔄 By midday, automated reports hit their inboxes and impact their progress trackers on the most critical workloads. They measure this against standard averages for the same day of week in previous months and track deviations. 📤 By evening, custom GPTs and Microsoft Graph APIs assemble and send tailored updates to their key stakeholders. They feed these into knowledge bases for upcoming weekly or monthly reports, and leverage reflective agents or simple conversations to review the day and journal their progress. But what’s most remarkable isn’t just the elevation of planning, automation and across-the-board "offloading". It’s how they use the time it frees up and "onload" an entirely new category of goals and deliverables, most of which couldn't even be dreamed of even a year ago. 🛠️ They redesign processes and run resilience analytics at scale (think “code reviews” for a manufacturing footprints). 👩‍🔬 They optimize decision-making by feeding aggregated expertise Deep Research reports to cross-functional process owners. 🤝 And they’re not just "getting trained", they’re actively training themselves and each other with and on AI, leveraging our platforms to constantly share best practices and explore new territories and ethical guardrails. Which reminds me of Karim Lakhani and Iavor Bojinov's awesome DSAIL program at Harvard Business School! This is why, as I read recent news reports about the "AI Job Apocalypse", I think it’s dangerous for leaders to evaluate new hires against AI based on outdated benchmarks. ♾️ The baseline is broken. We’re at a historical inflection point, like fresh graduates bringing laptops into a world still run on pen and paper, only exponentially more disruptive. 🙈 Dismissing junior new hires is how you miss a technology revolution. They are your number one change agents. You’ll never see the solution they come with, if you only view them as a problem to solve. AI’s impact on the future of work isn’t over-financed. It’s under-dreamed. Don't let youth on the sidelines, or you're never getting a second chance at dreaming it for your organization. Curious to hear from Jeremy Utley, Eric Porres, Carl Miller and anyone thinking about what’s next for new entrants to the workforce. #AI #FutureOfWork #JobApocalypse #NewHires #GrowthMindset

164

28 Comments

Barbarians at the Gate (https://lnkd.in/gkEBX79s) is a very interesting new paper from some great folks at Berkeley, with some exciting results showing the potential for AI in systems research. But I think the authors aren't quite asking the hardest problem about where this takes systems as a field. I wrote a new blog post about it: https://lnkd.in/gxS3s4v4 The core point of the paper is that AI can be effective at automatically driving the 'number go up' part of systems research. Given an evaluator, AI can automate progress on implementation. I believe this is 100% correct. The paper also talks about how more focus will have to go into developing evaluators. Again, I 100% agree, but think they're understating their case. Building evaluators (basically tests) is, I believe, going to become the focus of systems and software over the next decade. Which leaves researchers with the hard problem of choosing which problems to go after. That's never been easy in academic systems, and is only going to become more of a challenge. But the best systems research is driven by choosing problems well! This vision for AI could lead to lots more of the least interesting systems research. The "number go up" kind. Which could make it harder than ever to identify the truly innovative and visionary work. We're already seeing that elsewhere in science. So, mostly, I think the paper is right, but not yet asking the biggest question: what should systems research be?

276

12 Comments

A quiet holiday week in the VC/AI world but still an important week for some interesting companies in the space. Cerebrium - Seed - $8.5m - Cerebrium is an AI infrastructure platform that allows engineering teams to build, deploy, and scale multimodal AI applications. Fastino - Seed - Undisclosed - Fastino enable developers everywhere to do more with AI. GrowthBook - Venture - Series Unknown - $22.6m - GrowthBook is an open source feature flagging and experimentation platform Cekura (YC F24) - Seed - $2.4m - Cekura AI is an AI based Software Development firm offering Developer Tool to observe and test AI voice agents. Manta - Venture - Series Unknown - $2m - Manta is the AI platform reinventing how organizations access and apply data science. As always if I missed anyone please let me know.

10

The most recent Queued Up report from Berkeley Lab is out. Steven Zhang and I are proud to be listed as coauthors on this edition. The Interconnection.fyi team worked closely with Joseph Rand and the rest of the LBNL team to provide the data that powers this year's analysis. Supporting this level of research is core to our mission of bringing transparency to the wholesale energy markets. The 2025 edition (covering data through the end of 2024) highlights some significant shifts in the landscape: Active Capacity: 2024 closed with nearly 2,300 GW of generation and storage seeking interconnection. Changing Mix: Active natural gas capacity increased by 72% year over year, while solar and storage saw slight decreases in total queue volume. The Backlog: 408 GW of capacity already has an executed or draft interconnection agreement but has not yet reached commercial operations. Timelines: For projects built between 2018 and 2024, the median duration from request to operation has doubled compared to the early 2000s. This report is the definitive annual benchmark for the industry and provides a vital baseline as we begin to see the implementation of FERC Order 2023. The analysis in this report is based on our EOY 2024 data snapshot. Since then, we have continued to track and update these queues every single day. If you want a live view of how these numbers have shifted in the months since this snapshot was taken, follow Interconnection.fyi and subscribe to our Substack. You can find the full slide deck, interactive maps, and raw data files at the link in the comments.

41

3 Comments

Viam begins where silicon ends and motion begins. Founded in September 2020 by Eliot Horowitz, longtime CTO of MongoDB and Brown-trained computer scientist, Viam delivers on a vision he's chased since ShopWiki: to treat robotics software like modern web-app frameworks. He didn’t just launch a startup; he built an abstraction layer for the physical world, packing data, AI, and hardware into a unified engineering stack. Across that stack stands Simone Kalmakis, VP of Engineering, who stitched modular drivers, edge-cloud orchestration, and #ML inference into a system designed for real saying yes to unreasonable requests. Alongside her is Edouard Servan-Schreiber, VP of Solution Engineering, forging customer integrations across industries that actually work, not just pitch decks. Advisors like Carolyn Everson (Permira, ex-Instacart/Facebook) and Albert Wenger (Union Square Ventures) bring the kind of M&A and scaling fluency that turns engineering into category power. Viam's platform lets companies plug in #devicecontrol, #datacapture, #fleet-wide management, #predictivemaintenance, even #remotedebugging, without months of glue code. Their modular driver model and edge-cloud flexibility mean factory floors, arenas, yachts, QSR kitchens, even sonar fishfinders get full-stack intelligence with swap-in modules and reusable community code. Funding tells the story. Seed through Series C, Viam has raised $117M, including $45M in Series B (led by USV and Battery Ventures, Mar 2024) and $30M in Series C (led again by USV with Battery and Neurone, Mar 2025). That capital fuels real deployments: robotic sanding on Viking Yachts, AI-powered sonar for Kongsberg Discovery, fan-flow optimization and interactive demos at UBS Arena with the New York Islanders. Viam’s traction spans #robotics, #marine, #climatetech, #QSRautomation, #industrial manufacturing, and #sports. They’ve cut time-to-production from months to days. That’s the kind of shift that signals a new category: physical intelligence. Culture is engineered for impact: collaboration, curiosity, accountability, diversity. #Opensource isn’t lip service, it’s alive in the Viam Registry, driven by community-built modules. Hardware, software, ML engineers build shoulder-to-shoulder. Partnerships with WAVS and Billion Oyster Project reinforce the build-for-good DNA. Now hiring across #engineering, #solutionengineering, #productmarketing, #projectcoordination. If you thrive under ambiguity, this is your sandbox, and stakes are high. Viam is writing software in the real world, across #yachts, #arenas, #kitchens, #boats, where every inch matters. Follow Viam's journey, explore roles, book a demo, and let's see how much reality you can tame. #Startups #StartupSpotlight #StartupGrowth #FounderStory #TechEcosystem #StartupEcosystem #Data #DataDriven #AI #Hardware #AIHardware #Cloud #CloudComputing #Technology #Innovation #Hiring #TechHiring If engineering peace of mind is what you crave, Vention is your zen.

29

2 Comments

Meet BeTaL: an LLM-in-the-loop framework that automates benchmark design. LLMs reason, simulate, and refine to create dynamic benchmarks that evolve as fast as the models they test—2–4× more precise than traditional methods. 📄 Research: https://lnkd.in/gRqAYtd9

62

1 Comment

The next big shift in user interfaces is already underway. While everyone now uses ChatGPT, the full power of these LLMs is showcased with Claude's native use of MCP (coming soon to ChatGPT as well). In my latest blog post, I explore how AI, Model Context Protocol (MCP), and stablecoins are reshaping the way we interact with digital services. Imagine a world where your AI agent handles everything - no more app fatigue, no more tapping through endless menus. The future is UI-free, conversation-driven, and powered by programmable money. What does this mean for developers, users, and the future of commerce? Check out the full post for a deeper dive. https://lnkd.in/ejZM6KPz

17

One of the surprising realities of working with large language models (LLM) is that you can feed in the same prompt and still get different outputs, even when you set temperature = 0. LLMs are fundamentally probabilistic, and while lowering the temperature reduces randomness, it doesn’t guarantee determinism. With greedy decoding, subtle factors like floating-point precision, parallel computation, tie-breaking logic, and model architecture quirks (e.g. Mixture-of-Experts) can cause outputs to diverge across runs. The result: temperature = 0 increases consistency, but doesn’t ensure it. For applications where reproducibility matters, it’s important to understand where this variability comes from and how to mitigate it. If you’ve ever been puzzled by the output differences and want to know why, give it a read and let me know what you think. https://lnkd.in/enpsRH_V

39

3 Comments

AI is just search. Love this new post by 🤖 Jacky Liang on why: - Semantic search fails where precision matters - Vector databases are doomed - Claude Code’s grep-powered search destroys Cursor’s vectors - Hybrid search (combining lexical, full-text, and semantic) is the future My favorite line: "Similarity ≠ relevance. Vector search can’t find getUserById." If you’re building an AI app, whether it’s a code assistant, chatbot, or product search, this piece is a must-read. And this is just the beginning. We’re kicking off a series showing how to: - Build multi-search systems using Postgres - Combine surgical precision of lexical search with fuzzy power of full-text - Add vector search for semantic understanding (when it actually helps) Read Jacky's post here: https://lnkd.in/g5wrUDbp

69

5 Comments

Here’s a truth that often gets missed: becoming an AI-first company at the enterprise level is still really hard. It requires making big investments in R&D and specialized talent and betting on still-evolving frameworks and tools — and having the commitment to push the field forward yourself where the right tools or integrations don’t yet exist for your specific use case. As deploying AI and agents becomes table stakes in more industries, some companies end up hyping prototypes or easy use cases on earnings rather than building complex, self-improving AI systems to move the needle on productivity and profit. The task of separating these AI talkers from AI doers in public markets  – and leveraging AI to do it – is why my longtime partner John Hughes and I created the Swift AI Index as a fun side project last year. We just updated the index with five new stocks as part of a quarterly refresh; check it out: https://lnkd.in/eNBGjQzK

79

7 Comments