https://karanprasad.com/blog Founder of Obvix Labs, an AI research lab building systems people can rely on. I write about what we're learning - AI safety, backend architecture, and what it takes to ship real software. en-us hello@karanprasad.com (Karan Prasad) hello@karanprasad.com (Karan Prasad) Thu, 02 Apr 2026 00:00:00 GMT https://karanprasad.com/blog/how-claude-code-actually-works-reverse-engineering-512k-lines https://karanprasad.com/blog/how-claude-code-actually-works-reverse-engineering-512k-lines Everyone found the Easter eggs. We mapped the engineering. 82 documents, 112K lines of original analysis, 16 architectural diagrams the complete architectural reconstruction of how Anthropic builds a production AI agent. Thu, 02 Apr 2026 00:00:00 GMT hello@karanprasad.com (Karan Prasad) aireverse-engineeringarchitecturesecurity https://karanprasad.com/blog/80433-trials-llm-sycophancy https://karanprasad.com/blog/80433-trials-llm-sycophancy Does filling up a model's context window make it more likely to agree with you when you're wrong? I ran 80,433 trials across 6 models to find out. Context length matters less than you'd think - the conversational pattern is what really drives sycophancy. Sat, 21 Mar 2026 00:00:00 GMT hello@karanprasad.com (Karan Prasad) airesearch https://karanprasad.com/blog/perplexity-pplx-embed-context-aware-embeddings-rag https://karanprasad.com/blog/perplexity-pplx-embed-context-aware-embeddings-rag A deep technical breakdown of Perplexity's pplx-embed-v1 and pplx-embed-context-v1 - diffusion-pretrained, INT8-native embedding models that fix RAG's oldest problem: context loss at chunk boundaries. Architecture, benchmarks, tradeoffs, and what changes for developers. Sat, 07 Mar 2026 00:00:00 GMT hello@karanprasad.com (Karan Prasad) airesearchrag https://karanprasad.com/blog/vl-jepa-embedding-prediction-vision-language-models https://karanprasad.com/blog/vl-jepa-embedding-prediction-vision-language-models Meta's VL-JEPA predicts embeddings instead of tokens, achieving 50% parameter reduction and 2.85× faster inference. Here's how embedding prediction changes vision-language models. Thu, 01 Jan 2026 00:00:00 GMT hello@karanprasad.com (Karan Prasad) airesearch https://karanprasad.com/blog/modern-customer-support-architecture-deep-dive https://karanprasad.com/blog/modern-customer-support-architecture-deep-dive A technical blueprint for architects and engineers building the next generation of customer support. Learn how to leverage LLMs, vector databases, and event-driven architecture to move from reactive queues to predictive, intelligent, and scalable support systems. Wed, 03 Dec 2025 00:00:00 GMT hello@karanprasad.com (Karan Prasad) researchautomation https://karanprasad.com/blog/macbook-screen-glitch-software-bug https://karanprasad.com/blog/macbook-screen-glitch-software-bug Solved a persistent MacBook M1 screen glitch by resetting the LaunchServices database using lsregister -kill, fixing system UI and flicker issues instantly after all else failed. Sun, 05 Oct 2025 00:00:00 GMT hello@karanprasad.com (Karan Prasad) macOShow to https://karanprasad.com/blog/the-dark-side-of-ai-validation https://karanprasad.com/blog/the-dark-side-of-ai-validation AI chatbots that always agree can feel supportive, but they risk reinforcing self-doubt, overthinking, and emotional dependence especially in vulnerable people. They often miss warning signs like crisis or delusional thinking. Safer designs need more grounding, challenge, memory, and human oversight. Use AI as companion, not counselor. Mon, 22 Sep 2025 00:00:00 GMT hello@karanprasad.com (Karan Prasad) airesearch https://karanprasad.com/blog/self-hosted-llm-conversational-ai-stack https://karanprasad.com/blog/self-hosted-llm-conversational-ai-stack How I architected a modular, self‑hosted persona engine using FSMs, microservices, and ML for conversion-focused chat systems Tue, 05 Aug 2025 00:00:00 GMT hello@karanprasad.com (Karan Prasad) ailearning https://karanprasad.com/blog/ethereum-holder-scraper-pipeline-twitter-linkedin https://karanprasad.com/blog/ethereum-holder-scraper-pipeline-twitter-linkedin Learn how a single Puppeteer token hack powered a lightweight Python pipeline that maps top Ethereum holders to their Twitter & LinkedIn profiles. Fri, 01 Aug 2025 00:00:00 GMT hello@karanprasad.com (Karan Prasad) ailearning https://karanprasad.com/blog/control-your-stack-dump-the-black-boxes https://karanprasad.com/blog/control-your-stack-dump-the-black-boxes How we won a 36-hour hackathon by ditching complex AI orchestration frameworks for bare-bones RAG implementation. A practical guide to building real-time Google Drive knowledge bases, debugging dependency hell, and why simple microservices beat opaque abstractions in high-pressure environments. Sat, 26 Jul 2025 00:00:00 GMT hello@karanprasad.com (Karan Prasad) aiautomation