Adding Live Web Search to a Local LLM Without Making the Terminal Wait
How nishad.ai added LLM-routed Exa Lite search, progressive local-first answers, Cloudflare secret isolation, SmolLM fallback support, and a measured 10-prompt latency benchmark.
Notes from the main branch
Technical blog
Ideas, implementation details, and lessons from building software.
Inside nishad.ai: Ontology Routing, Context Windows, and Two-Slot KV Caching
A technical walkthrough of the structured ontology, two-stage source routing, 4,096-token prompt budget, Qwen memory footprint, transformer KV caching, measured two-slot latency gains, and the next steps for nishad.ai inference.
Running nishad.ai on a MacBook: llama.cpp, a Python Gateway, and a Browser Fallback
How nishad.ai moved its default AI path to a Mac-hosted Qwen model using llama.cpp, an authenticated Python streaming gateway, Cloudflare routing, launchd recovery, privacy-safe inference metrics, and a WebGPU fallback.
Two Ways We Turned a MacBook Air into a Gaming Machine
A technical field guide to running Counter-Strike: Condition Zero and Counter-Strike: Source through a paid CrossOver setup, then streaming GTA V from a PS5 to a MacBook Air and a 4K monitor with PS Remote Play.
One Hour, Five Fixes, and One Missing Contract
A production debugging story about Spotify metadata, Gemma 4, SSE versus JSON, honest completion signals, Firefox PDF downloads, and the tiny UI details that decide whether a system feels trustworthy.
Inside the nishad.ai Terminal: A Self-Hosted AI Shell with Gemma 4
An implementable high-level design of the browser shell, knowledge pipeline, Gemma 4 inference stack, streaming gateway, security boundary, MLOps workflow, and future RAG path behind nishad.ai.
Building a Privacy-First Spotify Mood Recommender with Local Embeddings
How I combined Spotify history, Liked Songs, ephemeral private-message embeddings, grounded song metadata, vector ranking, and one-minute OAuth health checks.
Building nishad.ai: A Static Website with a Terminal, an AI Persona, and a Markdown Blog
Static Markdown blog publishing: JSON frontmatter, draft validation and exclusion, permalinks, builds and GitHub Pages deployment, alongside nishad.ai's vanilla homepage and shared AI ontology.
Build a Privacy-First OpenClaw Image Stream
A twelve-chapter tutorial covering OpenClaw, privacy-safe image search, four async lanes, sqlite-vec caching, Kafka recovery, and Azure migration.
From GitHub Pages to Private Analytics: Building My Own Metrics Dashboard
A metrics-only Go and PostgreSQL dashboard beside a static website: explicit consent, daily browser estimates, owner-only GitHub sign-in, and honest deployment boundaries.