The Datacenter and Beyond
VL2, FaRM, eRPC, Spanner, and CausalSim. The datacenter papers I don't remember, the ones I do, and what ten weeks of CS244C changed about how I think.
AI/ML insights and consulting
VL2, FaRM, eRPC, Spanner, and CausalSim. The datacenter papers I don't remember, the ones I do, and what ten weeks of CS244C changed about how I think.
Congestion control is half of computer systems. Four papers from Stanford's CS244C on fairness, adversarial networks, and why overprovisioning is the honest answer.
Two-Phase Commit, Zookeeper, Raft, and Simplex. Four papers on consensus from Stanford's CS244C, and why mathematical perfection matters less than defense in depth.
End-to-End Arguments, Tussle in Cyberspace, and Wallbleed. Three papers from Stanford's CS244C that reframed how I think about where intelligence belongs in a system.
190 controlled A/B tests show MCP costs 1.2-2x more tokens than CLI tools, but tool gaps -- not protocol overhead -- cause the real failures.
Every Tesla data service had a dealbreaker. So I built a streaming telemetry pipeline from scratch, with Claude doing the heavy lifting. Here's the decision process, the architecture, and what the data actually looks like.
56 hand-labeled entries said Azure OpenAI was the best embedding model. 7,813 entries with cross-validation say otherwise -- and reveal why F1 = 0.46 isn't a failure.
How I compared 4 embedding models for entity disambiguation in a Cosmere knowledge graph -- and what the metrics actually mean.
AI adoption and potential gain are inversely correlated -- the people who would benefit most aren't using the tools. Here's the research, the math, and what to do about it.
A walkthrough of setting up remote access to Claude Code from an iPhone using Tailscale, mosh, and tmux, including the detours, dead ends, and the surprisingly simple setup I ended up with.
I tested 9 frontier AI models with a deceptively simple question. The results reveal a fascinating tension between statistical priors and reasoning ability.
Nanny Timesheet, Short Projects, Courses
Courses, OpenClaw, Short Projects, Personal Projects
courses, short-projects, openclaw
An interactive map of 32,000 home sales in King and Snohomish County, with traffic-aware commute times to work and climbing gyms.
short-projects, courses, bhavanaai
courses, short-projects, bhavanaai
Analyzing 1,234 GitHub incidents from 2013-2026 to test whether rising outages correlate with the Azure migration or product expansion.
Analyzing daily PM2.5 and AQI data from 277 Indian cities to find that the most polluted places are small towns, not the metros that make headlines.
courses, openclaw, tools, bhavanaai blog, short projects
courses, personal-finance, Azure cleanup, openclaw
personal-finance, courses
personal-finance, courses, openclaw
courses, career, openclaw, personal-finance
courses, tools, openclaw, personal-finance
tools, openclaw, courses, personal-finance
courses, tools, openclaw, personal-finance
openclaw, tools, courses, personal-finance
courses, personal-finance, tools, openclaw
tools, openclaw, courses, personal-finance
courses, tools, openclaw
courses, tools, openclaw
courses, tools, openclaw
courses, career
courses
courses, claude-transcript-viewer, podcast-summarizer-v2
courses, podcast-summarizer-v2, claude-transcript-viewer
tools, podcast-summarizer-v2, claude-transcript-viewer, courses
tools, podcast-summarizer-v2
claude-code-transcripts, courses-stanford-cs244c, bhavanaai, tools
podcast-summarizer-v2, claude-code-transcripts, courses-stanford-cs244c-gavel
podcast-summarizer-v2, courses, tools
courses
podcast-summarizer-v2, courses, tools
courses, podcast-summarizer-v2, tools
podcast-summarizer-v2, career, courses
podcast-summarizer-v2, career, courses
courses
CS244 GPU scheduling traces, GFW censorship research
CS244 project research