Skip to content

Latest commit

 

History

History
123 lines (91 loc) · 5.93 KB

File metadata and controls

123 lines (91 loc) · 5.93 KB

aimake — Adoption & Improvement Roadmap

Ideas to drive faster adoption, retention, and production use. Prioritized by impact and effort.


Adopt immediately (high impact, clear hook)

# Feature Why it matters
1 aimake init --from generators ✅ Done — makefile, dvc, prefect, airflow-dag
2 One-command pipeline templates aimake init --template=rag|finetune|eval-only|agent with working example — clone and run in 2 minutes
3 VS Code / Cursor extension ✅ Done — extension/ sidebar plan (stale/reuse/restore), status-bar cost, build/explain
4 Official GitHub Action ✅ Done — .github/actions/aimake with cache + PR comment
5 Remote cache that “just works” aimake cache remote init — cloud bucket setup in ~30 seconds (not only manual S3 YAML)
6 Cost dashboard in CLI aimake cost / aimake plan --cost as a first-class story — see the bill before you pay
7 LangChain / LlamaIndex / HF adapters ✅ Done — docs/ADAPTERS.md
8 aimake watch ✅ Done — aimake watch [--build]
9 Better “why stale?” UX ✅ Done — aimake explain --tree, --format json
10 Comparison docs ✅ Done — docs/COMPARISON.md

Strong adoption (team & production)

# Feature Why it matters
11 Web UI (lightweight) ✅ Done — Next.js + Tailwind dashboard + aimake serve / aimake graph --serve
12 Shared team cache ✅ Done — team_id, cache remote-init, pull-lock, lock v2 remote pin
13 Artifact registry with remote ✅ Done — S3 / HF / W&B push + registry push
14 Policy / approval gates ✅ Done — policy.promote gates on CLI + dashboard
15 Scheduled builds ✅ Done — aimake schedule + schedule.jobs
16 Notifications ✅ Done — Slack / Discord / email + notify-test
17 Multi-project monorepo ✅ Done — --project / -P
18 Secrets integration ✅ Done — .env + Vault / Doppler / 1Password

Trust & correctness (why people stay)

# Feature Why it matters
19 Auto-detect external model drift ✅ Done — external.probe + aimake probe
20 Custom validation scripts ✅ Done — validation.command
21 Output signing / attestation ✅ Done — SLSA-lite sidecars via attestation.enabled
22 Reproducibility report ✅ Done — aimake repro (+ dashboard)
23 Partial restore from cache ✅ Done — missing outputs → RESTORE on cache hit
24 Lineage export ✅ Done — OpenLineage / MLflow / W&B + dashboard

Developer experience & ecosystem

# Feature Why it matters
25 Docker image ✅ Done — Dockerfile + GHCR publish (ghcr.io/.../aimake)
26 Plugin marketplace / entry points Third-party plugins without editing core (pip install aimake-wandb-plus)
27 aimake doctor --fix Auto-fix missing dirs, suggest yaml fixes, detect stale partial outputs
28 Interactive TUI ✅ Done — aimake tui Rich full-screen plan/build/metrics
29 Jupyter integration %aimake build evaluation in notebooks — huge for researchers
30 TypeScript / Python SDK parity ✅ Done — aimake.sdk + @aimake/sdk + docs/SDK.md

Growth & distribution

# Feature Why it matters
31 Benchmark / savings page “Prompt change: 45 min → 90 sec, $12 → $0.40” with reproducible example
32 Content & listings Dev.to / YouTube 5-min demo, Awesome-LLM-Ops list
33 GPU host integrations Modal, Replicate, RunPod, Lambda — “run GPU step on X” as one yaml block
34 Free hosted cache for OSS Sign in with GitHub → 10GB cache — lower barrier vs rolling your own S3
35 Compatibility badge README badge showing cache hit rate / last build status
36 Documentation website ✅ Done — Next.js + Tailwind docs site in website/ (port 3001)

Top 5 — highest adoption per engineering week

Priority Feature Rationale
1 GitHub Action + PR cost/quality comment Every ML team uses GitHub CI
2 aimake init --template=rag Instant “it works” moment
3 Market aimake plan cost + tokens Already shipped in v1.2.0 — lead with this in docs and demos
4 LangChain / LlamaIndex one-liner integration Meet existing code, don’t replace it
5 aimake watch + Cursor extension ✅ Extension shipped in extension/; watch already done

Strategic positioning

aimake’s wedge is incremental + cost-aware + AI-shaped — not “another orchestrator.”

Tool Strength
Airflow / Prefect Scheduling and orchestration at scale
DVC Data versioning
Make Generic file dependencies
aimake Skip unchanged AI pipeline steps; show cost before run; prompt/model/config-aware

Double down on:

  1. Cost before run
  2. Skip what didn’t change
  3. 5-minute template to first win

Already shipped (reference)

  • Incremental builds, content fingerprints, parallel execution
  • S3 remote cache, GPU scheduling, distributed workers
  • Experiment compare, hyperparameter optimization, artifact registry
  • Plugins: Hugging Face, W&B, DVC, Docker, Ollama
  • v1.2.0: external dependencies, atomic outputs, output validation, plan cost estimates

See CHANGELOG.md for full release history.


Contributing

Pick an item, open an issue, or PR with the feature number from this doc (e.g. “Implements #8: aimake watch”).