aiollm
An async-first, typed HTTP client toolkit for AI APIs: OpenAI and Anthropic clients, validated responses, retries, uploads, and streaming. Currently pre-alpha.
View repository ↗Python 3.10+ / backend / distributed systems
Backend & Distributed Systems Engineer specializing in typed, async-first Python.
Seven years in Python. From API contracts to production operations.
@GrehBan / GitHubprojects
Python libraries and developer tools I build and maintain. Open a repository for the source code and current project status.
An async-first, typed HTTP client toolkit for AI APIs: OpenAI and Anthropic clients, validated responses, retries, uploads, and streaming. Currently pre-alpha.
View repository ↗An async callback executor with bounded concurrency, configurable result ordering, and managed task lifetimes.
View repository ↗Structured exceptions with typed fields, message templates, and context that can redact sensitive values.
View repository ↗An experimental generator for Pydantic models and aiollm API methods from OpenAPI 3.0/3.1 specifications.
View repository ↗Strictly typed data structures for HTTP headers, authentication, cookies, and timeouts, with RFC validation.
View repository ↗A bilingual product site for aiollm, with developer documentation, usage examples, and project information.
View repository ↗focus
Typed APIs, data systems, messaging, infrastructure, testing, and observability. The right combination depends on the system.
Typed Python APIs, asynchronous services, and outbound HTTP integrations.
Relational, document, cache, and time-series storage with deliberate schema evolution.
Queues, streams, and task processing chosen for the delivery model each system needs.
Linux-first delivery, container orchestration, and web-server operations.
Metrics, traces, and logs that make distributed services easier to operate.
Fast feedback, realistic infrastructure in tests, and property-based edge-case discovery.
Repeatable checks and delivery pipelines for Python services.
Async API tooling for AI/LLM services, networking, and Telegram automation.
about
Seven years in Python, working across the full backend lifecycle: typed API boundaries, async networking, persistence and migrations, distributed work, delivery, and production visibility.
FastAPI and Starlette with Pydantic cover HTTP services; asyncio, aiohttp, and HTTPX support async work and integrations. The data layer spans PostgreSQL, SQLAlchemy 2.0, Alembic, MongoDB, Redis, InfluxDB, and direct SQL—different tools for different data and access patterns.
RabbitMQ quorum queues and NATS JetStream offer distinct messaging models; Celery handles task workflows. I choose between them based on delivery guarantees, replay needs, and operational fit—not by combining every broker by default.
Docker, Docker Swarm, Nginx, Apache, Linux, Bash, and GitHub Actions take services toward production. Pytest, pytest-asyncio, Testcontainers, and Hypothesis support confidence; Prometheus, Grafana, OpenTelemetry, the LGTM stack, and Loki help make runtime behavior visible.
blog
Notes on Python, async systems, typing, and the small decisions that shape dependable tools.
A few design choices that make Python API clients predictable without hiding the HTTP layer.
↗Why task ownership and cancellation boundaries matter in async Python services.
↗How parsing and validation help keep internal Python code simpler and safer.
↗contact
Have a Python API, distributed workload, or async integration to make more dependable? Send a note.