AI and infra engineer. Writing about AI agents, infrastructure, and building in public.
Two generic detectors failed on my content. So I built my own on my own ground truth: pre-2026 posts are human, 2026 posts are machine-written. It scored 44 for 44. Then I attacked it for an evening, drove its verdict on this very post down to 0.313, and packaged the attack as a skill.
How I built Spin Lab, an interactive table-tennis spin course, by keeping the physics pure, precomputing every shot into a frame timeline, and asserting the prose itself in a headless test suite. Plus the calibration bug that proved four of my serves were physically impossible.
Notes on agent loops, deterministic workflows, Claude Fable, OpenRouter Fusion, and Google Open Knowledge Format.
AI-assisted notes from my liked-tweets feed, organized around agent loops, cloud agent infrastructure, skill security, memory, and runtime context.
Notes on agent engineering: what counts as an agent, why long tasks fail, harness design, eval loops, observability, and when teams build vs buy.
AI-assisted notes from my liked-tweets feed, organized around agent workflows, browser traces, model loops, and guardrails.
One WhatsApp thread with Hermes Agent captures what I do every 15 minutes and what I eat, writes two JSON files, and this site renders two public readouts from them. Why it works, and why I only track two things.
AI-assisted notes from my liked-tweets feed, organized around harnesses, managed agents, memory, workflow packaging, and enterprise AI work.
High thinking budgets can reduce control, especially when stale context and autonomous tool use mix together.
One skills repo, two laptops, five agent runtimes. I let an agent design the sync, and every unstated decision gave it room to optimize for control instead of maintainability.
IAM policy rightsizing agent with AI-powered risk signals and instant rollback.
Local-first archive for AI chat traces across agents, designed as a substrate for DSPy/GEPA-style optimization loops.
Frontend-only LLM ensemble interface with bring-your-own OpenRouter/API keys and client-side multi-model workflows.
Ebook/document normalizer with optional OCR and LLM-generated advice + flowcharts. It also became a practical OCR evaluation project for me: for most books, plain PDF text extraction worked better than heavy OCR, so GLM OCR ended up as a fallback reserved for noisy pages, scans, or extraction failures.
MCP server for hashline-anchored safe file read/edit workflows inspired by this archived write-up: https://web.archive.org/web/20260412162933/https://blog.can.ac/2026/02/12/the-harness-problem/
Knowledge-graph representation of the Indian Income Tax Act so agents can traverse section-to-section relationships explicitly.
log deduplication sidecar using Drain3 algorithm, reduces volumes ~90%.
Wispr Flow OSS - live streaming voice-to-text using whisper.cpp with Metal acceleration.
Rust network diagnostics toolkit I built while debugging inconsistent Wi‑Fi and suspected service-quality issues on my own machine; the goal was to see what was actually happening under the hood instead of relying on vague network status indicators.
Open-source attempt to replicate Gemini-style long-form YouTube understanding: extract MP3 audio with ffmpeg, build timestamped transcripts, detect only meaningful frame changes, and feed the useful video/audio context into an LLM. I found the visual path added little value, so it effectively converged into an audio-first podcast/video understanding pipeline.
Kubernetes CronJob controller with Prometheus gating, pacing, and jitter.
Watch your codebase breathe; TUI for real-time file change monitoring with inline diffs
offline Wikipedia with hybrid no‑index RAG - powered by gpt-oss:20b
I wanted one cleanup workflow after each feature and tried building a Pi extension for it. Packing all three steps into a single prompt failed quietly, so the extension now runs them as sequential gated steps.
The written version of my talk on agentic design. The model is one layer; the interesting system is the harness, context, memory, control loop, and review process around it.
A shared mount looked like the cleanest way to make two agent instances talk across machines. A 30 minute directory cache made it unreliable, so Git became the bus instead.
I mounted an S3 bucket as a local folder in Finder with rclone and FUSE-T, auto-mounted on login. This is the full setup, the flags that make it behave, and where it falls over.
Once I had LLMs editing code autonomously, I wanted to watch a directory breathe in real time. Existing tools gave me file events or diffs or a TUI, never all three, so I built vigil.
A prompt rule for running parallel agents without conflicts: the coordinator owns the shared work, sub-agents only get isolated slices.
How a nostalgia trip through Windows XP on UTM failed, why QEMU could not provide the Direct3D support Chessmaster needed, and why Whisky ended up being the actual fix.
eBPF CPU scheduler enforcing latency SLOs via Earliest Deadline First.
Decision audit ledger for AI agents - captures "why" as queryable data.
You can run AWS Batch jobs as pods in your Kubernetes cluster. Pretty cool actually. As expected the AWS tutorial is reliable, you follow it step by step...
How is pod level CPU usage metric calculated in k8s? How is pod level CPU usage metric calculated in k8s? The common PromQL query found on the internet may...
Handy script to download all your business from dropbox This script had to undergo a lot of iterations, it failed a lot of times after download for hours on...
Note: Some steps (public accessibility, IAMFullAccess, blob handling) are simplified/dated. Modern migrations avoid public endpoints, use fine grained IAM,...
Note: This post reflects my 2019 exploration of geofence validation. Today, libraries like Turf.js or GEOS handle these cases more efficiently. Suppose you...
I can write about the whole incident in detail because this was my own startup. I was the one man tech team here for about 2 years. What happened = 20th Feb...
Let the service name be ‘banana’ Let the pods be of the service ‘pineapple’ There was a rise in 502s accompanied by cannot lookup banana and interestingly...
Kubernetes: TIL in a k8s pod when you define env vars, if you define the same env var twice, the latter one takes precendence how resource quotas work in...
Reminder: you can just do things a new word pedagogy. which means the method of teaching something. another unrelated word tacit knowledge. knowledge that...
When does one scale? When one hits a bottleneck. That’s when.What is scaled in that scenario?Only the instance running the service which has become the...
I was reading about how secondary indexes work in redis. Some people think that well hey, redis is a key value store isn’t it? So why would someone wanna...
What a joy! It was a breath of fresh air when I came across the requirement of implementing live streaming. It was quite a learning curve to understand how...
OpenTelemetry auto-instrumentation worked in a Python REPL but not in the actual FastAPI server. The reason was PEX running Python with -sE flags, which ignore PYTHONPATH. How I found it, and two ways to fix it.
Explaining CPU, RAM, and disk to juniors with a kitchen analogy: chef, counter, pantry. Covers iowait, steal time, throttling, overcommit, and IOPS vs bandwidth.
VPS, VM, and containers get used interchangeably. They solve the same problem - packing more workloads onto fewer machines - at different abstraction levels. The breakdown I give juniors.
There is nothing more that needs to be said.
This is what seems to be working for me and appears to be sustainable My goals are fat loss and muscle retention for now Legit been trying to lose weight...
Kinda depressing shit tbh I found these notes below from 2019, I was an inspired kid. 2025 today I am overwhelmed, I don't dream as freely as I used to. A...