Balaji Krishnamurthy

Writing The Markup every week forces clarity. On our portfolio. On private markets. On where frontier tech is really moving. It’s work. AI helps with the volume, but the thinking is ours. If you want to see what we’re watching in real time, here’s this week’s edition.

30

Qsi Bhairav is the pause between events. Not mystical silence, but a lawful checkpoint in system design. In our architectures, QSi Bhairav is: • The neutral audit state → tamper-proof buffers where inputs are captured, hashed, and time-stamped. • The synchronization anchor → quantum-secure timing that ensures fidelity across every packet and surveillance event. • The immutable ledgering gap → the lawful intermission where evidence is notarized before state change. This “pause” is not absence—it is pure attention. The middleware of trust. The verification enclave regulators can rely on. In regulated ecosystems, Bhairav becomes the backbone: transparent, audit-ready, and immune to manipulation. #QuantumSecurity #BlockchainEvidence #SurveillanceInfrastructure #AuditReady #RegTech #SystemArchitecture #Innovation #TrustByDesign #QuantumTiming #ComplianceTech #DeepTech #Governance

2

Overview The Unleashing AI Innovation in Financial Services Act (H.R. 4801 in the House and S. 2528 in the Senate) is a bipartisan bill introduced on July 30, 2025. It aims to accelerate AI adoption in U.S. financial services by creating a regulated framework for testing AI solutions without risking immediate enforcement action. Key Provisions 1. Creation of AI Innovation Labs (Regulatory Sandboxes) Federal financial regulators—including the Federal Reserve, FDIC, OCC, SEC, CFPB, NCUA, and FHFA—must establish AI Innovation Labs to allow financial institutions to pilot AI projects in a controlled environment without facing enforcement actions, provided eligibility requirements are met. 2. Application Process for AI Test Projects Financial institutions must apply through their regulators, describing the AI project, any requested regulatory waivers, and proposed alternative compliance methods. 3. Eligibility Criteria Projects must serve the public interest (including consumer protection), improve operational efficiency or competitiveness, and pose no systemic risk to the U.S. financial system. 4. Bipartisan Support and Strategic Alignment The Act has strong bipartisan backing and aligns with federal goals to balance AI innovation with robust consumer safeguards. Summary Table Component Details What Establish AI Innovation Labs (regulatory sandboxes) at financial regulatory agencies Who Agencies: Fed, FDIC, OCC, SEC, CFPB, NCUA, FHFA Why Enable AI experimentation while maintaining consumer protection and system safety How to Participate Submit application with project details and waiver proposals Key Requirements Public interest, efficiency gains, no systemic risk

40

2 Comments

Looking back, it all began with one conversation. A candid discussion with a friend who was working as an analyst at D.E. Shaw at the time, opened my eyes to how hedge funds think about supply chain transparency. 'Tracking where cargo moves and combining it with satellite data gives them an edge worth millions.' That 15-minute conversation turned into a months-long deep dive for us. The more we researched, the clearer it became that tracking alone wasn’t enough. We uncovered a bigger gap! Hedge funds don’t just want to track commodities. They aim to detect what others can’t - sanctioned cargo, shifting trade flows, and global chokepoints. That challenge is exactly where PierSight steps in. Our latest blog explores how SAR + AIS data gives hedge funds 24×7 visibility — enabling them to act on supply chain intelligence and build alpha. https://lnkd.in/gU58i56k

55

4 Comments

🚀 AI Experiment – AI Without LLM (…for now) TL;DR: I added an “AI chatbot” to my static portfolio website without an LLM — it retrieves facts but doesn’t reason or converse. Adding an LLM will make it context-aware, human-friendly, and action-ready. I recently built an “AI chatbot” on thatvishalkumar.dev that runs RAG without the “G-Generation”. Here’s how it works in simple terms: 1️⃣ Created a Vector DB based on my personal data in the codebase 2️⃣ Query comes in – e.g., “What’s Vishal’s current role?” 3️⃣ RAG/Vector search compares the query against my stored data (think of it as numbers representing each sentence) 4️⃣ Top snippet retrieved – e.g., “Senior Vice President - Tech at 360 One.” 5️⃣ Answer returned – the system simply outputs that snippet ✅ Works for fact retrieval ❌ No reasoning, no rephrasing, no context 🤔 So, what’s an LLM and where does it come in?: Large Language Models are not databases of truth. Think of them as smart language translators: Understand your query in human terms Look at retrieved knowledge Generate a natural, conversational answer Remember context across multiple turns 🏢 How This Scales to Enterprise Apps: Existing DB → Vector DB: convert structured or unstructured data into vectors so it can be searched semantically RAG retrieval: fetch the most relevant context for any query LLM generation: turn snippets into readable, human-friendly answers MCP (Multi-Channel Processing): trigger actions like updating dashboards, sending emails, or scheduling meetings 💡 Key Takeaways: RAG + Vector DB = factual accuracy LLM = conversational intelligence MCP = actionable insights Next step: soon i will wire in an LLM and share the before/after differences.

28