Writing
Blog
15 posts
The Machine That Passed the Medical Exam but Cannot Feel a Handshake
I showed a demo to a client last month. The AI answered every question correctly. Then he asked me - does it actually understand what it's saying? I paused. Honest answer is no. This post is about why.
Claude is Dreaming: How Anthropic Borrowed a Billion Years of Evolution to Fix AI Memory
Anthropic just shipped a feature called Dreaming for Claude agents. I had read the exact same idea three weeks ago in a book about the human brain. Here is what is actually happening and why it matters.
AI Silicon Strategy 2026: What the Hardware Wars Mean for Your Cloud Spend
OpenAI, Google, Anthropic, and China's Four Dragons are betting billions on different silicon strategies — here's what it means for which cloud you should trust with your AI workloads.
Stop Wasting 80% of Your LLM Tokens: The Caveman + Graphify Framework
A three-step framework — compressed prompting, knowledge graph indexing, and project-specific shorthand — that cuts token usage by 80-90% without sacrificing technical accuracy.
AI-Powered Code Review at Scale: Building a Multi-Provider VS Code Extension
How I built a bring-your-own-AI code review VS Code extension that found 12 critical security issues in my own codebase — including SQL injection risks I'd missed for months.
ModelRouter: Intelligent Multi-LLM Routing to Stop Burning Claude Pro Tokens
How I built a local proxy that classifies prompts and routes them to the right model — Gemma for trivial tasks, Gemini for features, Codex for debugging, Claude only for hard architecture problems.
The Context Gap: Why Frontier LLMs Fail at Local Business Problems
Frontier LLMs are trained on Western-centric internet data and fail at regional supply chains, local business etiquette, and jurisdiction-specific law — here's how localized fine-tuning fixes this on commodity hardware.
MCP as Infrastructure Control Plane: Building a Real-Time Monitoring Dashboard in 2 Hours
How I used a lightweight MCP shell server to build a real-time infrastructure monitoring dashboard — and why MCP is more than just an AI agent protocol.
Beyond RAG: Building AI Systems That Actually Know Your Business
Why vanilla RAG isn't enough, and how the four-layer architecture — memory, identity, tools, and feedback loops — builds AI that genuinely represents how your organization thinks.
Building a Multilingual Meeting Notes System for Indian Business: OdiaMeet
How I built a system that transcribes meetings in 10 Indian languages including code-mixed conversations, and generates clean English minutes of meeting at near-zero cost.
WhatsApp as Enterprise Interface: Conversational AI for Business Automation
Why WhatsApp beats native apps for customer-facing business automation, and the architecture pattern that makes it work across bookings, support, onboarding, and more.
Building Production AI Agents with MCP and Anthropic Claude
How I built a production-grade AI agent using Model Context Protocol (MCP) connected directly to a client's production database — delivering real-time natural language data intelligence.
Apache Solr as a NoSQL Database: When Search Beats Storage
After running 150+ Solr nodes in production at TransUnion CIBIL under RBI regulation, here's when Solr outperforms MongoDB and when it doesn't.
Graphify: Temporal Knowledge Graphs for Stateful AI Agents
How I built Graphify — a tool that turns any folder of code, docs, or notes into a navigable knowledge graph — and why temporal memory is the missing piece in most AI agent architectures.
Zero-Downtime Apache Solr in Production: VIP + Keepalived Architecture
How we achieved zero-downtime patching and version upgrades across a 150+ node Solr cluster serving a national credit reporting platform under RBI regulatory constraints.