What I work with day to day. Updated September 2026.
Terminal
- tmux (prefix
C-Space, resurrect + continuum, fzf popups). One session per ticket, windows for edit / run / AI, one window per cluster whose name picks the kubeconfig. - zsh with starship, fzf + fd + bat + ripgrep, direnv, mise, zoxide.
- Neovim (LazyVim) as the editor. lazygit for git, k9s for clusters.
- Everyday CLI:
jq,yq,curl,mc(MinIO client),skopeo,sops+age,keyringfor secrets in scripts,ghandglab,caffeinate. - Aliases that earn their keep:
gst,gdf,gru,ggpush,ggpull,kctx,kns,ktop,twin(tmux neww -n).
Kubernetes and infra
- kubectl, kubectx/kubens, helm, k9s, Argo CD, GitLab CI.
- Local clusters: kind and k3d (OrbStack), kwok for scheduler experiments, a fake GPU operator for GPU scheduling without GPUs, skaffold for build-and-load loops.
- Terraform and Ansible for what lives outside the cluster.
Languages and tooling
- Python: uv, pytest, ruff, black, mypy. Flask and FastAPI services, Celery on RabbitMQ.
- Go: controllers, operators, exporters (client-go informers).
- Rust: cargo, for CLI tools and anything that must be fast and small.
- Bash and Jinja: glue scripts, templated manifests and configs.
- Node.js / TypeScript when the job calls for it; Erlang/OTP in a former life (XMPP).
Observability
- Prometheus and Thanos (multi-cluster), Grafana, Loki, NVIDIA DCGM and AMD SMI exporters.
AI tooling
- Claude Code in the terminal, with my own skills: Clean Architecture and Clean Code review lenses, a career knowledge base that refreshes itself, a contributions ledger mined from git, graphify for codebase knowledge graphs.
- Things I built for this workflow: gitgossip (commit messages and repo digests from git history) and tmux-claude-attn (flags tmux windows where Claude is waiting on you).
- The posts here come out of a small content engine that indexes my repos, ranks stories worth telling, and drafts them with an evidence bundle. That will get its own post.
Hardware
- MacBook Pro 14" M2 Pro, macOS.
- MacBook Pro 16" M3 Pro, macOS.
- NVIDIA DGX Spark.
Languages
- Bengali (native), English (fluent), Arabic (beginner).
Books that shaped how I work
- Clean Architecture and Clean Code, Robert C. Martin
- Designing Distributed Systems, Brendan Burns
- Building Microservices, Sam Newman
- System Design Interview, Alex Xu
- Learn You Some Erlang, Fred Hébert
- Head First Design Patterns, Freeman, Bates, Sierra, Robson