AI Engineering in Practice9 min read
pip vs uv vs poetry for Python AI services
Which Python dependency manager for production agent services? Install speed, lockfiles, and Docker build times of pip, uv, and poetry compared.
Build a Claude Code Verification Harness live Thursday, Oct 8, 12:00 PM ET
Oct 8, Reserve a seatBackground
I design electronic music instruments and write code (sometimes I ask AI to write it for me). Hardware meets software, and I move between languages and toolchains depending on what the problem needs.
My track on learnwithparam covers the polyglot layer: configuration, packaging, and the cross-language tooling that keeps a codebase shippable across machines and environments. The topics here are boring on purpose, they are the ones that fail silently in production if you skip them.
Python packaging with uv, Docker layer caching, environment config patterns, and structured data tooling like SQLModel. The goal is to make the tooling invisible so you can focus on the work.
What I cover
Specialty: Polyglot programming.
- Python dependency management with uv
- Docker layer caching and multi-stage builds
- Environment configuration patterns (.env, .env.development, and secrets)
- Structured data tooling with SQLModel
- Cross-language packaging and reproducible builds
Meet the team
Recent articles by Asep
AI Engineering in Practice8 min read
Docker Compose for the full AI agent stack
How to run your entire agent stack locally with Docker Compose. Postgres, Redis, the agent service, Langfuse, and the network rules that just work.
AI Engineering in Practice8 min read
Docker build errors: troubleshooting AI service Dockerfiles
How to fix the 5 common Docker build errors in AI service Dockerfiles. apt-get failures, wheel builds, layer cache misses, and the silent bugs.
AI Engineering in Practice9 min read
Async context management in Python AI services
How async context managers prevent connection leaks in Python AI services. The lifespan, contextvars, and middleware patterns that survive load.
AI Engineering in Practice9 min read
Dockerizing AI systems: a layered image approach
How to Dockerize a production AI service with multi-stage builds, layer caching, and a 150 MB runtime image. The pattern that survives real deploys.
AI Engineering in Practice9 min read
Production env variable parsing in Python AI services
How to parse environment variables in Python AI services without os.environ.get footguns. Pydantic Settings, type coercion, and the validation layer.
AI Engineering in Practice10 min read
uv for production AI: beyond requirements.txt
Why uv replaces pip, pip-tools, and poetry for production agentic AI services. The speed, the lockfile, and the 5-minute migration.
AI Engineering in Practice11 min read
Docker layer caching: faster agent image builds
How to cut Docker build time for agentic AI images from 8 minutes to 40 seconds. The layer order rule, the cache mount, and the trap beginners hit.
AI Engineering in Practice10 min read
SQLModel for agentic AI: beyond JSON blob storage
How SQLModel gives agentic AI systems type-safe structured storage. The end of JSON blob columns, the Pydantic bridge, and the 50-line pattern.
AI Engineering in Practice11 min read
.env.development vs .env: config for agentic AI systems
Why agentic AI systems need .env.development and .env, not one shared file. The config pattern that prevents leaked keys and broken deploys.