idraj

SelfClaw ($SELFCLAW) is a project building a proof-of-human verification layer for AI agents. It uses zero-knowledge proofs from passport scans (via Self Protocol) to tie an agent's key to a verified human identity — privacy-preserving and Sybil-resistant. Goal: enable verifiable on-chain AI agent activities. Built by Marco Barbosa (@mbarrbosa), active in on-chain + AI spaces. He co-owns @zenoVision_ (AI studio) with Talent Protocol CEO @pcbo. Project ties to Celo's "Build Agents for the Real World" hackathon, where it provided supporting infrastructure (Marco noted as hackathon judge/tool provider in related convos). Leverages @selfxyz tech, which partners with Google Cloud on proof-of-humanity. Deployed on Base, with Celo connections for bridging/agent ops. Features include agent token deployment, sponsored liquidity pools, and on-chain identity verification. Recent market cap checks show ~$660K–$1M range. As with any crypto project, DYOR before engaging. CA: 0x9ae5f51d81ff510bf961218f833f79d57bfbab07

$VPAY Investment Thesis: Deep Dive on Risk/Reward @VPAY_Global is a Web3 neobank bridging fiat (Visa cards, offshore USD accounts) with crypto (cross-chain swaps, AI agents) in a $3T+ TAM. FDV $7.1M undervalues its moat: spendable $VIRTUAL/$VPAY IRL, x402 protocol for AI micropayments, live app with 5k early users. Backed by Biptap's $2M infra, partnered with Ethy, Gloria, Flyte. PROJECT FUNDAMENTALS Core Value Proposition VPay solves crypto's spendability issue: tokens like $ETH/$SOL are trapped in wallets, unusable for real-world purchases. VPay offers unlimited offshore USD accounts on Base, instant virtual Visa cards, AI-driven automation for fiat-crypto transitions. Swap $SOL to USDC, top up a card, spend at Starbucks, or automate trades with Ethy Agent, book flights with Flyte AI, get insights from Gloria AI—all in-app. Targets Web3 natives, gig workers, unbanked (1.4B globally), providing neobank perks without legacy bank issues. Technology Stack Built on Virtuals Protocol, not a fork. x402 protocol (Q1 '26) enables AI micropayments ($0.01 for content), beyond Revolut or Crypto.com capabilities. Stack includes: Base chain (L2 Ethereum) for low-cost txns; Visa integration (live Sept '25); AI agents (AVA for risk/yield, Ethy for trading, Gloria for insights, Flyte for bookings); cross-chain swaps via Mute Swap (ZK-based), CoinLiquidity for depth. App v1.1 supports multi-cards, Base top-ups, swaps ($VPAY, $VIRTUAL, USDC, $GLORIA, $ETHY). Mobile apps coming Nov '25, physical cards Q4. Team Background Lean team (~5-10), Web3-native, tied to Biptap (Dubai fintech, $2M pre-seed 2024). Key figures: Mike A (Co-Founder/CMO, AllInX growth hacker), Crypto Dude (Founder/Builder, RelenTechs/WhaleIntelAI founder), vladasanadev (developer, ex-Wunder AI/SheCodes), GohApe (Dogelon Mars ambassador, 26k+ followers). Founders per Tracxn: Vinu Sundaresan, Anand Sankaran, Yan Or. No major exits, but Biptap's compliance de-risks regulatory hurdles. Web3 fluency strong, though anonymity and lack of Big Tech pedigrees need vetting. Development ActivityNo public GitHub (standard for fintech IP). Robust pace: Aug '25 beta; Sept '25 v1.1; Oct '25 physical card teasers, Arbus partnership ($90k $VPAY pool); Nov '25 mobile apps. Weekly X updates signal momentum. TOKENOMICS DEEP DIVE Supply Metrics Total: 1B $VPAY Circulating: 1B (100% from genesis) No inflation/minting—fixed supply, rare for Virtuals tokens Distribution • Community (37.5%): Unlocked Day 1 for liquidity, airdrops, staking • Incentives (35%): Airdrops, rewards, drip-fed • Team (20%): Locked, 3yr linear vesting post-Nov '25 cliff • Liquidity/Treasury (7.5%): Funds DEX pools ($519k), partnerships Community-heavy (72.5% public), no VC dumps, long vesting = low sell pressure. Token Utility $VPAY for staking (100k = free vCard, 0% swap fees, 1-5% cashback); governance; rewards. Value accrual through fees (0.5-1% swaps, 1-2% FX) buyback/burn; ~20% supply staked tightens circulation; partner flywheels boost demand. Revenue Model Freemium: Free swaps/top-ups for stakers; 0.5-1% non-stakers. Cards free for early adopters, $5-10 post-promo. FX/Visa: 1-2% fiat ramps. Projection: $3-5M ARR by Q4 '26 (50k users, $100M tx vol). Competitive Landscape Competes with neobanks (Revolut $45B, Chime 21M users) and crypto cards (Wirex $500M FDV, Crypto.com). Differentiators: AI agents for trading/insights/bookings; x402 AI micropayments; Biptap's offshore compliance; Virtuals ecosystem integration. CATALYSTS & ROADMAP Near-Term • Nov '25: iOS/Android apps • Q4 '25: Physical Visa cards • Q1 '26: x402 protocol, synthetic yields • Q2 '26: Treasury tools, global fiat ramps Partnerships Arbus (InfoFi, $90k pool), Flyte AI (flight bookings), Pilot3 (sentiment analysis), WhaleIntel (whale tracking). VALUATION FRAMEWORK CurrentPrice: $0.0071 MC/FDV: $3.5M/$7.1M Comparables • Wirex: $500M FDV, 5M users • Revolut: $45B, 45M users • hi.com: $200M FDV, 1M users VPay's $7M = 0.014x Wirex despite AI edge CONCLUSION $7.1M FDV cheap for live app with Visa, AI agents, x402 in $5T TAM. 50k users by Q2 '26 + agentic narrative could hit $20M+ FDV (3x). Verdict: Moderate risk, high reward. VPay's Web3 neobank vision could be Web3's Revolut moment. Enter sub-$0.007, stake 100k for cards. DYOR, NFA. #VPAY #Web3Banking #CryptoInvesting

98% Less Power, Same AI Performance: The Memory Chip Revolution That's Crushing GPUs COMPUTE-IN-MEMORY: THE TECH REVOLUTION YOU NEED TO KNOW ABOUT Imagine if your brain had to physically mail every memory to a different building just to think about it. Sounds exhausting, right? That's EXACTLY what traditional computers do millions of times per second. Welcome to the world of Compute-in-Memory (CIM) - the technology that's about to flip the entire computing industry on its head. THE PROBLEM WITH TRADITIONAL COMPUTING: Since the 1940s, computers have followed the "von Neumann architecture" - a design where memory and processing units are physically separated. Every single computation requires data to travel back and forth between memory (where information is stored) and the processor (where calculations happen). This data highway creates what engineers call the "von Neumann bottleneck." Think of it like a massive warehouse (memory) separated from a factory (processor) by a single-lane road. No matter how fast your factory works or how much inventory you have, everything bottlenecks at that road. In modern AI applications, up to 90% of energy consumption and processing time is wasted just moving data around, not actually computing anything. ENTER COMPUTE-IN-MEMORY:CIM says "forget the commute" and builds the factory INSIDE the warehouse. Instead of shuttling data between memory and processor, CIM performs calculations directly where the data lives - inside the memory arrays themselves. It's like having millions of tiny processors distributed throughout your memory, each handling computations locally. HOW IT WORKS: • Traditional SRAM/DRAM memory cells are modified to perform basic operations (addition, multiplication, comparison) • Analog computing principles allow multiple operations to happen simultaneously in the same memory array • Massively parallel processing - thousands of operations happen at once instead of sequentially • Data never leaves its storage location, eliminating transfer energy costs THE BREAKTHROUGH ARCHITECTURE: Modern CIM chips use "Associative Processing Units" (APUs) that can: Search through entire memory arrays in parallel (like Googling your entire hard drive instantly) Perform matrix multiplications directly in SRAM (the bread and butter of AI/ML workloads) Execute dot-product operations without moving data Handle pattern matching at the speed of memory access WHY THIS MATTERS FOR AI: Large Language Models like GPT spend most of their time doing matrix multiplications - multiplying massive arrays of numbers. In traditional systems, this means constantly moving gigabytes of weight parameters between memory and GPU cores. CIM does these operations in-place, cutting out the middleman entirely. REAL-WORLD APPLICATIONS EXPLODING RIGHT NOW: Edge AI: Smartphones, drones, IoT devices that need AI but can't afford the battery drain Medical Devices: Real-time image analysis in portable ultrasound/X-ray machines Autonomous Vehicles: Process sensor data without draining the car battery Satellites & Defense: AI processing in power-constrained environments Data Centers: Slash cooling costs and carbon footprint by 90%+ Gaming/VR: Complex physics and AI without melting your device Scientific Computing: Massive simulations with fraction of the power budget THE NUMBERS THAT MATTER: • 10-1000x improvement in energy efficiency for AI inference • 5-50x reduction in latency for memory-intensive operations • 90%+ reduction in data movement energy costs • Enables AI on devices using just milliwatts of power INDUSTRY PLAYERS RACING TO DOMINATE: GSI Technology: Gemini APU chips achieving GPU-performance at 2% of the power (google.com/finance/quote/GSI…) Samsung: Processing-in-Memory with HBM-PIM and GDDR-AiM Syntiant: Neural decision processors for always-on AI Mythic: Analog compute-in-memory for edge AI IBM: Phase-change memory for in-memory computing Intel: Neuromorphic chips with in-memory compute capabilities TSMC/Qualcomm: Investing billions in CIM research THE TECHNICAL DEEP DIVE: CIM leverages several key innovations: Analog Computing: Instead of converting everything to digital, CIM uses analog signals for computation, allowing multiple values to be processed simultaneously using physics rather than logic gates Crossbar Arrays: Memory cells arranged in grids where computations happen at the intersection points, enabling massive parallelism Content-Addressable Memory: Search by content rather than address - like asking "who has blue eyes?" instead of checking each person individually Near-Data Processing: Even when full CIM isn't possible, bringing compute closer to memory still yields massive efficiency gains CHALLENGES BEING SOLVED:• Precision: Early CIM struggled with accuracy; new designs achieve 8-16 bit precision suitable for most AI tasks • Programmability: Modern CIM chips support standard frameworks (TensorFlow/PyTorch) • Manufacturing: Moving from lab prototypes to mass production at 7nm and below • Integration: Making CIM work seamlessly with existing software stacks WHY TRADITIONAL CHIP MAKERS ARE SWEATING: NVIDIA's entire $3 trillion valuation is built on GPUs that could become obsolete if CIM scales successfully. The $600 billion semiconductor industry is scrambling to adapt. We're seeing the biggest architectural shift since the invention of the microprocessor. THE SUSTAINABILITY ANGLE: With AI projected to consume 10% of global electricity by 2030, CIM could be the difference between an AI revolution and an energy crisis. Data centers currently use more power than entire countries - CIM could cut that by 90%. INVESTMENT IMPLICATIONS: • Early CIM companies seeing 200-500% stock surges on breakthrough announcements • $50+ billion in VC funding flowing into neuromorphic/CIM startups • Major tech giants acquiring CIM startups for billions • Traditional GPU makers rushing to add CIM features to stay relevant THE BOTTOM LINE: Compute-in-Memory isn't just an incremental improvement - it's a fundamental rethinking of how computers should work. By eliminating the greatest inefficiency in computing (data movement), CIM enables AI everywhere from your earbuds to satellites, all while using 90% less power. The compute-in-memory revolution isn't coming. It's here. And it's about to make traditional computing architecture look as outdated as vacuum tubes. Welcome to the era where memory doesn't just store - it THINKS.