https://omnihash.com/ monthly 1.0 https://omnihash.com/work/ai-agent-platform-0-to-1 yearly 0.8 https://omnihash.com/work/crypto-rpc-cost-audit yearly 0.8 https://omnihash.com/work/healthtech-clinical-search yearly 0.8 https://omnihash.com/blog 2026-06-29 weekly 0.9 https://omnihash.com/blog/local-ai-development-gpu-cost 2026-06-29 monthly 0.7 https://omnihash.com/images/diagrams/gpu-prices.png The true cost of a local AI rig once you add the machine around the card: a 3090 build near $2,300, a 4090 build near $3,600, a 5090 build near $5,300, and an RTX PRO 6000 workstation near $17,000. On the cheaper builds the GPU is barely half the total. https://omnihash.com/images/diagrams/gpu-price-rise.png GPU prices have climbed over roughly the last two years, the reverse of the usual slide in used-hardware prices. The two newest cards use their launch price as the baseline. https://omnihash.com/images/diagrams/cost-over-time.png Cumulative cost over two years, counting the whole rig: a one-time 3090 build (around $2,300) as a near-flat line against $20, $100, and $200-a-month plans, with break-even near twelve and twenty-three months. https://omnihash.com/blog/true-cost-of-an-employee-token-spend 2026-06-26 monthly 0.7 https://omnihash.com/images/diagrams/cost-equation.png The cost of an employee in the AI era: salary plus benefits plus token spend. https://omnihash.com/images/diagrams/cost-stack.png Two hires, total cost against useful output. The cheap hire has a lower salary but a large, mostly wasted token-spend segment and little useful output; the quality hire has a higher salary, a small efficient token-spend segment, and high output. https://omnihash.com/blog/llama-cpp-vs-ollama 2026-06-22 monthly 0.7 https://omnihash.com/images/diagrams/llamacpp-ollama.png Two paths to one engine. Ollama is the easy, opinionated route (one-line pulls, a model library, a REST API); llama.cpp is the control-and-performance route (GGUF, quantization, raw kernels). Both converge on the same GGUF model running on your GPU. https://omnihash.com/blog/nextjs-vs-vite-vs-remix 2026-06-16 monthly 0.7 https://omnihash.com/images/diagrams/vite-composable.png A bundled framework fuses build, router, rendering, and data into one block you adopt whole. A composable setup keeps them as separate layers you choose and swap independently. https://omnihash.com/blog/pulumi-vs-iac-ease-of-use 2026-06-09 monthly 0.7 https://omnihash.com/images/diagrams/iac-spectrum.png The infrastructure-as-code spectrum, from declarative YAML through a domain-specific language to a general-purpose programming language. Tools sit where their authoring model lives, not where their power ceiling is. https://omnihash.com/blog/mcp-explained-tools-auth 2026-05-28 monthly 0.7 https://omnihash.com/images/diagrams/mcp-architecture.png The MCP shape. A host application with an MCP client connects across a trust and auth boundary to one or more MCP servers, each backed by a real system. Identity and scopes flow down the same boundary. https://omnihash.com/blog/evals-for-llm-apps 2026-05-19 monthly 0.7 https://omnihash.com/images/diagrams/eval-gate.png The eval gate. Any change runs against the eval set, produces a score, and meets a decision: clear the bar and ship, miss it and loop back to iterate. https://omnihash.com/blog/agent-orchestration-and-delegation 2026-04-28 monthly 0.7 https://omnihash.com/images/diagrams/agent-orchestration.png The orchestrator-worker pattern. A coordinator fans work out to specialized sub-agents running in isolated contexts, then fans their results back into a single synthesis step. https://omnihash.com/blog/rag-chunking-strategies 2026-04-15 monthly 0.7 https://omnihash.com/images/diagrams/chunking-strategies.png The same document, split two ways. Left: fixed-size cuts land mid-idea and strand a fact from its context. Right: structural and semantic cuts fall on natural boundaries, so each chunk stands on its own. https://omnihash.com/blog/prototyping-with-disposable-html 2026-04-07 monthly 0.7 https://omnihash.com/images/diagrams/prototype-loop.png The disposable prototype loop. Iterate cheaply until a direction is validated, then branch out to a real build. The loop is the point; the files are not. https://omnihash.com/blog/oidc-github-actions-aws 2026-03-12 monthly 0.7 https://omnihash.com/images/diagrams/oidc-flow.png The token-for-credentials handshake. No secret is stored anywhere in the pipeline; the workflow proves who it is, and AWS hands back credentials that expire in minutes. https://omnihash.com/blog/why-ai-pilots-never-ship 2026-02-26 monthly 0.7 https://omnihash.com/images/diagrams/pilot-gap.png The demo-to-production gap. Most pilots reach an impressive demo, then fall into the gap that separates it from production. A thin bridge, scoping the pilot as a production slice, is the only thing that crosses. https://omnihash.com/blog/agent-driven-development-in-practice 2026-02-18 monthly 0.7 https://omnihash.com/images/diagrams/agent-loop.png Generation is fast; merging is not. The review bar does not drop. If anything, agent-driven development makes review more important, because volume goes up. https://omnihash.com/blog/build-vs-buy-ai 2026-01-22 monthly 0.7 https://omnihash.com/images/diagrams/build-buy.png Build for what differentiates you and what no vendor will fit. Buy everything else. https://omnihash.com/blog/rag-in-production-what-matters 2025-12-15 monthly 0.7 https://omnihash.com/images/diagrams/rag-pipeline.png A RAG pipeline. The quality of the answer is capped by the quality of retrieval. Most teams over-invest in the last box and under-invest in the first. https://omnihash.com/blog/ai-mvp-cost-and-timeline 2025-11-18 monthly 0.7 https://omnihash.com/images/diagrams/scope-first.png Scope-first sequencing. The MVP exists to test the riskiest assumption, not to be a small version of the final product. https://omnihash.com/blog/rag-vs-fine-tuning 2025-10-21 monthly 0.7 https://omnihash.com/images/diagrams/rag-vs-ft.png RAG supplies knowledge; fine-tuning shapes behavior. Most "the model does not know our stuff" problems are knowledge problems. https://omnihash.com/images/diagrams/rag-decision.png Start with the cheapest thing that could work. Prompting and a good eval set solve more problems than either technique alone. https://omnihash.com/blog/add-ai-to-existing-product 2025-09-23 monthly 0.7 https://omnihash.com/images/diagrams/value-risk.png Pick a first feature in the high-value, low-risk quadrant. Let trust, not enthusiasm, move you rightward. https://omnihash.com/images/diagrams/ai-boundary.png A single internal boundary keeps provider details, prompts, and guardrails in one place, and lets you swap or add models without touching the product. https://omnihash.com/blog/crypto-rpc-cost-audit-lessons 2025-08-19 monthly 0.7 https://omnihash.com/images/diagrams/cost-curve.png Inefficient per-action call patterns turn linear user growth into super-linear cost. The leak was always there; scale just made it visible.