ML/AI, distributed systems, data infrastructure
AI systems, data infrastructure, and engineering tradeoffs.
AI is moving fast, and coding agents are taking on more of the implementation work. As the work changes, so does what engineers need to learn and remember. I want to use this blog to write down the things I come across in practice that I think are worth knowing: core ideas, implementation details, and lessons from ML/AI, distributed systems, and data platforms.
Latest writing
-
OpenAI Agents SDK vs. LangChain DeepAgent: which one to use
A practical comparison of OpenAI Agents SDK and LangChain's Deep Agents, with brief looks at Claude Agent SDK and CrewAI.
-
How We Monitor Our Agentic Platform, Plus Datadog vs. Braintrust
How we built monitoring for our agentic platform with Datadog, and what we learned comparing it with Braintrust along the way.
-
Some Advanced Python Topics - Async and Parallel Jobs
Quick notes on asyncio, ThreadPoolExecutor, and ProcessPoolExecutor, with examples of when to use each.
-
Some Advanced Python Topics - Logging
Quick notes on logging beyond basicConfig: loggers, handlers, formatters, filters, and queue handling.
-
One Agent or a Team of Agents?
It feels obvious that a team of AI agents should beat a single one. Three recent studies say: usually not — and the exceptions are where it gets interesting.
-
Encoder vs. Decoder Today
Quick notes on a common AI interview question: what encoders and decoders do, when to use them, and what changed in the ChatGPT era.
-
Best Resources for Understanding Attention and Transformers
Best resources that helped engineers like me understand transformer attention, from the math intuition to the full architecture.
-
3-Min Summary: How OpenAI Scaled PostgreSQL to Millions of QPS
What I learned from OpenAI's PostgreSQL scaling post: protect the primary, scale reads, and keep operations disciplined.
