Andrei Panferov

A Founder should prepare a Data Room like this for easy access to their documents. Always remember that an organized and well-prepared Data Room shows you are professional, transparent, and ready to scale, which eventually speeds up the due diligence process. Some Points to note while preparing a Data Room: 1. Structure it clearly: use separate folders for different sections like Company Formation Documents, Financials, CapTable, Customer Details, Patents, etc. 2. Make sure that all documents are updated and accurate 3. For Security purpose, try to use a trusted platform with access controls 4. Only include necessary documents Here is a simple structure to get you started (see attached image for reference) 1. Company Summary: Pitch Deck, One Pager 2. Company Documents: Incorporation documents, Shareholders Agreements 3. Team: Contracts & ESOP, Advisors 4. Client Details: Contracts, Pipeline, Usage Data, Market Research 5. Legal: Technology & Security Agreements, Patents/ Trademarks, Insurance, Policies, Additional Agreements 6. Investment Documents- CapTable status, Investment Documents (SAFE, Convertible Note) 7. Financials- Budget, Financial Projections Credit: yana_vc, Pretiosum Ventures ~~~~~ ♻️ Found this helpful? Repost it so your network can learn from it, too. And follow me, Fazlur Shah, for more content like this. #startups #entrepreneurship #venturecapital #investing

167

15 kommentarer

We just launched a new LLM that was fine tuned to reduce false positives in watch-list screening. The headliner results beat all expectations - ◼️ 98% accuracy in name degradation testing (vs ~80% for others) ◼️ True positive to false positive ratio improved from 100:1 to 3:1 ◼️ 100% recall during testing across X PEP/sanctioned individuals ◼️ 20X reduction in false positives per 1000 checks At a high level, we trained the models to understand the context surrounding each potential hit. Then gave them thousands of historical examples of known analysts' resolutions to learn the difference between true positives and false positives. Our own Joshua Linn wrote a deep dive article explaining their process. Link in comments.

149

5 kommentarer

🚀 The EU AI Act is now law, and some deadlines have already passed. If you're building an AI startup, you might not realize that certain obligations kicked in on February 2, 2025 (prohibited AI) and August 2, 2025 (general-purpose AI models). The big wave hits August 2, 2026, when high-risk system requirements become mandatory. After diving deep into the Act's implications for startups, I've put together a comprehensive survival guide that cuts through the legal complexity. The reality? Your compliance burden depends entirely on your AI's risk classification, and most startups are overestimating which category they fall into. Here are three critical insights that might surprise you: 🤖 First, the Act includes genuine support for startups. You get access to free regulatory sandboxes with legal protection during testing, reduced penalties scaled to company size, and simplified documentation templates. The EU wants innovation, not strangulation. 🧭 Second, location offers no escape. Building in San Francisco or Singapore? If EU users can access your AI or its outputs reach the EU market, you're fully in scope. You'll need an EU representative and must meet identical requirements to European companies. 💻 Third, open-sourcing your AI model (if under 10^25 FLOPs) can exempt you from most obligations. Some startups are strategically going open-source to reduce compliance complexity while building community trust. The guide covers everything from risk classification tables to real cost estimates (one autonomous driving startup reported €344,000 per deployment, but typical high-risk systems need €10,000-50,000 initial investment), practical documentation templates, and a month-by-month action plan through August 2026. Whether you're building healthcare AI, HR tech, fintech, or chatbots, understanding these requirements now could save you from penalties up to €35 million or 7% of global revenue. More importantly, early compliance can become your competitive advantage as enterprise clients increasingly demand AI Act adherence. What's your take on AI regulation? Are you seeing it as a barrier or an opportunity to differentiate? #AIRegulation #EUAIAct #StartupCompliance #ResponsibleAI #AIGovernance

99

21 kommentarer

I've reviewed 1000+ AI healthcare pitches. The best founders can answer one question instantly: “Show me where your model was confidently wrong." Last week, a founder pitched their diagnostic AI. 97% accuracy. FDA breakthrough designation pending. Impressive. But when I asked about their model's failure modes, they went quiet. Red flag. The best AI healthcare founders I've backed obsess over edge cases: • One founder discovered their radiology AI had learned to detect the hospital's scanner model, not tumors • Another's "failed" patient risk model revealed hidden biases in training data that led to a better approach • A third pivoted from diagnosis to workflow optimization after their model kept flagging process inefficiencies In AI, understanding why models fail is more valuable than celebrating when they succeed. Now when I evaluate startups, I dig deeper into: 1) Their test set diversity 2) How they handle distribution shift 3) What happens when the model encounters data it's never seen Because in healthcare AI, the companies that design for failure build products doctors actually trust. What's the most surprising way you've seen an AI model fail that taught you something valuable? #HealthcareAI #HealthTech #VentureCapital #AIStartups #DigitalHealth #startups

119

26 kommentarer

Andreas Cleve (Corti) sat down with Bloomberg to unpack why general-purpose AI is falling short in healthcare—where nearly 80% of projects stall or never make it past the pilot phase. Meanwhile, a huge win for purpose-built AI: Corti is now being embedded even deeper into clinical workflows by Dedalus — the world’s third-largest EHR provider — bringing Corti’s technology to over 60 million patients across Europe and beyond. It’s a major validation of what specialized Gen-AI can do when it’s built for healthcare, not just adapted to it. Also, you can now explore Corti's API and apply for the Frontier Program — with credits, support, and early access to powerful new features. More details soon. Links in the comments.

88

4 kommentarer

We didn’t start MintNeuro to build just another neural implant. We started it because the chips inside them weren’t good enough. Thanks to Startups Magazine for the in-depth profile of MintNeuro and the journey we’re on: from world-class research labs to real clinical impact. At the heart of our work is a simple but urgent problem: today’s implantable devices rely on electronics that were never designed for the brain. We’re changing that, chip by chip. We believe precision, scalability, and clinical relevance start with the silicon. That’s why we’re creating purpose-built, ultra-efficient, modular chips for interfacing with the nervous system. Grateful to our partners for pushing boundaries with us – and to the funders, investors, programmes that are helping make it possible. Full article here: https://lnkd.in/e8tay-BG #semiconductors #neurotech #neuralimplants #neuralinterfaces #braincomputerinterfaces #bci #chips #deeptech #medtech

144

6 kommentarer

What is not an AI agent? That’s a crucial question in this world of AI hype. Here’s a useful Seussian poem that I wrote without AI help: It isn’t an agent if it looks like a chat, And an email “copilot” ain’t where its at It’s not web search or research alone, It’s not document rewrite in a pompous tone, Sure they’re AI, they’re cool, but agents they’re not, An agent is actually more like a bot, It makes its decisions, chooses what to do And it acts on its choices without an OK from you From Praveen Seshadri - an entrepreneur and ceo I respect

54

5 kommentarer

Who Owns the Code? IP, Equity & Co-Founder Showdowns We dive into a tough Reddit question from a technical founder getting squeezed out after building 100% of the product. • Set terms early: Always define equity and IP ownership before you start building. • No contract? No code. Without a signed IP agreement, the builder likely owns the product. • Don't get gaslit: If they say you're “not founder material,” remember—they picked you. Ask for fair equity—or walk. Don’t be afraid to say, “I’m out.” Hard truths, real stakes, and how to protect yourself as a founder.

41

6 kommentarer

🚀 Portfolio Highlight: Graphio 🚀 Founded by Dmitry and Andrew — repeat founders who previously built and exited fintech platform upSWOT (acquired by Uptiq.ai in 2024). Graphio.ai is building the AI-powered operating layer for CROs and COOs. It auto-generates SOPs from how work actually happens, compares it to how it should happen, and flags execution drift before it costs revenue. Why it matters? Because every missed renewal, silent pipeline stall, or broken handoff starts weeks earlier with invisible breakdowns. Graphio makes the invisible visible — turning what used to be an annual strategy project into a living operating system. We invested because we saw strong signals: ✨ Repeat founders who’ve scaled before. ✨ Backing from 100+ senior exec angels across Mastercard, PwC, Snowflake, Experian, HelloFresh, BambooHR and more — each committed to opening doors. ✨ A product wedge: 30-minute API onboarding vs. the heavy, months-long process-mining incumbents. ✨ And a category play: a new generation of execution compliance engine focused on assessing both humans and AI agents. Proud to back them on their journey. First Degree Capital 💜 Graphio.ai 👉 graphio.ai

135

37 kommentarer

Death by PoC (Proof of Concept): I recently met a promising startup. Great tech. Strong use case. They’ve been running multiple PoCs with major enterprises. But there’s a problem: No signed contracts. Just slow-moving maybes. This happens more often than we admit. Why? Because a PoC isn’t a win. It’s just the beginning of a sales process—often with no clear owner, no path to scale, and no urgency. Here’s how we can do better: Startups: • Don’t accept a PoC without success metrics and a post-PoC decision timeline • Map who holds the budget and who controls rollout early • Don’t pilot for “learning.” Pilot for a contract Enterprises: • Stop running pilots you don’t plan to scale (and I know you sometimes believe if it goes well, we will find a way to scale, but that attitude got so many startups bankrupted - get the internal buy-in and work on solving real problems) • Involve procurement before kickoff, not after results • Make it easier to say yes—or at least faster to say no! Founders: What’s helped you convert PoCs into real revenue? Operators: What makes you say “Let’s go” after a pilot? Let’s break the PoC trap—for everyone’s benefit. #CorporateInnovation #Startup

199

31 kommentarer

How much energy does Google’s AI use? One prompt = less than 9 seconds of watching TV and about five drops of water. In a recent study Google found that energy requirements for an average Gemini text prompt is substantially lower than many public estimates. AI workloads are in high demand, but there are a lot of work going into in reducing the power and water required per prompt. Important work that will only become more relevant over time.

37

15 kommentarer

Evals Evals Evals... the dire need for AI judges in medical applications. AI is rapidly expanding in biotech & healthcare but the complexity of use cases is outpacing the benchmarks we use to evaluate them. While models are getting better, the way we measure their performance hasn't kept up. Most benchmarks are too generic, outdated, or disconnected from real-world scenarios. As AI takes on higher-stakes roles — diagnosing conditions, guiding clinical decisions, automating hospital operations — this gap becomes a serious risk. Why this matters 👇 - Poor benchmarks = false confidence. A model that performs well on public datasets might fail in real settings due to unseen edge cases or biased data. - Slows down iteration. Without targeted evaluations, teams don’t know what to fix — and waste cycles on the wrong things. - Limits deployment. Without clear, high-resolution metrics, it’s hard to get regulatory buy-in or clinical trust from pharma, payors or healthcare providers. Here’s what I’d like to see built 👇 - Shift from generic to use-case-specific evaluations — tailor-made rubrics for tasks like pathology triage, patient-agent communication, or radiology interpretation. - More dynamic, real-time benchmarks — models judged continuously in production using hybrid AI + human scoring loops. - Expert-validated synthetic datasets — to fill gaps in real-world data, especially for underrepresented populations or rare edge cases. - Model evaluation as a feedback loop — with evaluation not as a static checkpoint but a tool that drives active data collection and fine-tuning (RLHF, DPO, etc). - Custom metrics > standard metrics — moving away from generic F1/Precision and building product-aligned scoring systems. AI benchmarks aren’t optional. They're the key to getting real-world AI adoption in medicine. Thoughts?

66

18 kommentarer

Frugal innovation. It's going to increasingly challenge US and European companies. DeepSeek AI just launched an AI product that caused a $1 Trillion collapse in stock prices today. With NVIDIA, Broadcom, TSMC and Oracle loosing over $700 Billion in market cap. Why? Frugal innovation. DeepSeek AI created an AI model that's at least as good if not better than US companies like OpenAI, Meta, Google have created, with only $6 million in funding. Compare that to the $21 Billion that OpenAI has raised. How? Necessity. Most technology companies around the globe don't have the luxury of receiving billions of dollars of Venture Capital. They have to grow frugally an focus on things like revenue and profits. Frugal Innovation is why I'm focused on technology companies in places like #Africa, #Latam, #MENA and #EasternEurope. Companies in those regions are by far much scrappier and focused on their customers and on business fundamentals vs US companies awash in VC. The don't have the luxury of ignoring business fundamentals. Another recent example. Nubank, a Brazilian all digital neo-bank currently has a market cap of about $60 Billion USD. That's more than the market cap of Barclays and Deutsche Bank combined. But, Nubank has only been in business for 12 years. Yes, US businesses will always have problems solved by Silicon Valley, and easy access to capital is often a moat and a competitivo advantage. But, increasingly, I believe we're going to see companies whose _lack_ of access to capital is a competitive advantage, and US companies should beware. Frugal innovation is a competitive advantage and a moat to other companies competing with you. Iyinoluwa Aboyeji Aman Verjee, CFA Dave McClure, Israel García Ballesteros Jose Kont Natasha Ibarra (She/Ella) Neil Shah Chris Mayo, CFA

39

6 kommentarer

The biggest mistake in AI right now is assuming things will move too slowly. Every founder and investor is trying to forecast the next 6–12 months of AI progress. But what you think will happen in 12 months might happen in 3. And errors of omission (missed upside) are much more dangerous than errors of commission (wrong bets). To reason about the future, we all build mental models: assumptions, beliefs, timing expectations. But exponential change breaks those models faster than we can update them. Claude Code now writes 95% of its own codebase. The best default assumption right now is that AI will advance faster than we expect. At Standard Metrics, we’ve already made the call: we are all in on AI. Starting with LLM-powered document parsing, and now extending to several soon-to-be-announced analytics and reporting features, AI is core to our strategy. The way we're operating our business (building, selling, writing, communicating) is also rapidly evolving as we adopt new AI tools. We'll be writing a lot about how AI is going to transform the private markets in the weeks and months to come. Let me know if there's anything specific you want us to address in future writing (or build toward in our product!).

60

8 kommentarer

Gaming founders, check this out 👇 A16Z GAMES is hosting the new cohort of Speedrun in SF. Speedrun is an early-stage accelerator for startups at the intersection of Tech and Games. The program is a complete curriculum on topics that define the category these days, so make sure you look at it and bookmark it. 👀 - Market Trends - Fundraising - People + Talent - Design + Infrastructure - Metrics + Revenue - Partnerships - Marketing + GTM - AI. They have invested in over 50 gaming startups across these sub-verticals:  - AI - AR/VR - Gaming studios - Games x consumer - Games x Web3 - Infrastructure and Tools - User-Generated Content (UGC). So, if you’re a gaming founder in the early stages, don’t miss out on what Andreessen Horowitz is doing in the space. They’re a foundational part of this emerging ecosystem, so watch what they do. Feel like you’re not ready to apply yet? Scroll through my content for inspiration. 😉 #gaming #founders #startups #accelerator #games #ar #vr

47

6 kommentarer