https://prabhavnalhe.com/notes/ Notes on distributed systems, reliability engineering, and LLM agents for infrastructure. en-us https://prabhavnalhe.com/notes/service-graph-lying/ https://prabhavnalhe.com/notes/service-graph-lying/ 05 Jan 2026 09:00:00 -0800 The dependency graph you draw from RPC traffic is real but incomplete. The dependencies that take you down are usually the ones no edge in that graph represents - shared config… https://prabhavnalhe.com/notes/streaming-ingest-rust/ https://prabhavnalhe.com/notes/streaming-ingest-rust/ 12 Jan 2026 09:00:00 -0800 When you fuse many noisy event streams into one model, the hard problems are not throughput - they are identity and time. Here is how I think about normalizing heterogeneous… https://prabhavnalhe.com/notes/closing-the-loop/ https://prabhavnalhe.com/notes/closing-the-loop/ 19 Jan 2026 09:00:00 -0800 A pattern worth reusing: when an LLM makes a judgment at scale, do not let it grade itself. Pair a cheap predictor with an expensive verifier that observes ground truth by… https://prabhavnalhe.com/notes/self-improving-platform/ https://prabhavnalhe.com/notes/self-improving-platform/ 26 Jan 2026 09:00:00 -0800 Most platforms that reason about a system are really stacks of queries against a model of that system. The recurring lesson is that the model is the constraint: build it as an… https://prabhavnalhe.com/notes/agent-harness-vs-rag/ https://prabhavnalhe.com/notes/agent-harness-vs-rag/ 02 Feb 2026 09:00:00 -0800 Two ways to feed a model context: retrieve fuzzy passages by similarity, or call purpose-built tools that return exact records. Here is how I decide which one a problem actually… https://prabhavnalhe.com/notes/evals-before-features/ https://prabhavnalhe.com/notes/evals-before-features/ 09 Feb 2026 09:00:00 -0800 Before you wire an LLM into a real workflow, decide how you will know it is good enough - because the eval is the gate, and the model is just the thing that has to pass it. https://prabhavnalhe.com/notes/llm-as-a-judge/ https://prabhavnalhe.com/notes/llm-as-a-judge/ 16 Feb 2026 09:00:00 -0800 A separate model can score outputs you cannot label by hand - but only if you treat it like a measuring instrument: a few binary dimensions, a calibrated reading, an answer key it… https://prabhavnalhe.com/notes/cost-aware-llm-pipelines/ https://prabhavnalhe.com/notes/cost-aware-llm-pipelines/ 23 Feb 2026 09:00:00 -0800 Most items in a large workload are easy, and a few are genuinely hard. The cheapest reliable pipeline is the one that spends accordingly: a deterministic fast path for the easy… https://prabhavnalhe.com/notes/apis-for-agents/ https://prabhavnalhe.com/notes/apis-for-agents/ 02 Mar 2026 09:00:00 -0800 When an LLM is the caller, the interface is the prompt. Typed responses, idempotent writes, granular composable tools, and machine-readable errors are not nice-to-haves - they are… https://prabhavnalhe.com/notes/delivery-is-the-product/ https://prabhavnalhe.com/notes/delivery-is-the-product/ 09 Mar 2026 09:00:00 -0800 A correct model or a sharp analysis is necessary but not sufficient. What gets internal AI actually used is the last mile almost no one designs for - routing each insight to its… https://prabhavnalhe.com/notes/slo-driven-risk/ https://prabhavnalhe.com/notes/slo-driven-risk/ 16 Mar 2026 09:00:00 -0800 Reliability effort gets spent on whatever feels scary in the room. Here is how to replace that gut feel with a defensible number built from SLOs, error budgets, and a clean… https://prabhavnalhe.com/notes/scaling-live-to-a-billion/ https://prabhavnalhe.com/notes/scaling-live-to-a-billion/ 23 Mar 2026 09:00:00 -0800 A live broadcast turns one source into millions of simultaneous viewers in seconds. The hard part is not the video - it is keeping a viral spike from melting your origin. Here are… https://prabhavnalhe.com/notes/text-to-sql-then-and-now/ https://prabhavnalhe.com/notes/text-to-sql-then-and-now/ 30 Mar 2026 09:00:00 -0800 In 2018 I helped build an RNN model that turned English questions into SQL, trained on WikiSQL and later published. Today a general-purpose LLM does the same thing with no…