IdeaWell

Project ideas from Hacker News discussions.

After 7 years in production, Scarf has reluctantly moved away from Haskell

Original Article

Hacker News Discussion

📝 Discussion Summary (Click to expand)

3 Dominant Themes in the Discussion

Theme Supporting Quote
1️⃣ Compilation speed is a deal‑breaker for LLM‑driven development “The number of compiler runs doesn't matter as much as the total elapsed time it takes to finish the task. In just about every test we ran, LLMs are faster at building in Python than Haskell.” — aviaviavi
2️⃣ LLMs perform better with constrained (less expressive) type systems “I've found LLMs to be best with more constrained type systems: they are better at ocaml than they are at typescript.” — giraffe_lady
3️⃣ Practically, teams are moving to simpler, faster languages (e.g., Python) even at the cost of strong typing “I am increasingly wondering if we are in a post‑language world in terms of development. Why would I ask an agent to write a server in anything other than the most efficient language, although efficiency can take several forms: runtime, token usage during development, and wall‑clock dev time (affected by slow compile times for example).” — matt2000

🚀 Project Ideas

Generating project ideas…

Haskell Incremental Compiler Accelerator

Summary

  • A cloud‑based incremental build service that caches GHC compilation output and only recompiles changed modules, cutting cold‑start times from minutes to seconds for LLM‑driven Haskell development.
  • Enables agents to iterate on Haskell code at near‑interpreted speed while retaining full type‑checking guarantees.

Details

Key Value
Target Audience Haskell developers using LLMs or agentic coding workflows who are blocked by 15‑minute cold compile cycles.
Core Feature Remote incremental compiler that tracks module dependencies, caches object files, and returns type‑checked binaries within seconds.
Tech Stack Haskell (GHC + Halfs), Docker/Kubernetes, Redis for dependency graph caching, GraphQL API, client libraries in Haskell & JavaScript.
Difficulty Medium
Monetization Revenue-ready: Subscription per build‑minute

Notes

  • HN commenters repeatedly cite “15‑minute cold builds” as the primary reason they abandon Haskell for faster languages.
  • A fast‑feedback sandbox would let agents converge on correct code without waiting for full recompiles, directly addressing the pain point.

Python Type‑Inference & Lint Companion for LLM‑Generated Code

Summary

  • A low‑friction VS Code/Neovim extension that auto‑generates precise type annotations and unifies disparate static‑analysis tools for Python, reducing tool sprawl for LLM‑generated code.
  • Provides a single, trustworthy type‑checking layer that integrates seamlessly with LLM output.

Details

Key Value
Target Audience Python developers who rely on LLMs and currently juggle multiple type‑checking tools (mypy, pyright, ruff, etc.).
Core Feature Auto‑generation of py.typed stubs, contextual type‑inference suggestions, and unified lint feedback displayed in‑IDE.
Tech Stack Python back‑end (mypy + typeshed), fine‑tuned CodeBERT model for type suggestion, Rust‑based CLI, Electron UI.
Difficulty Low-Medium
Monetization Hobby

Notes

  • Users such as mjr00 and kreyenborgi lament needing “2‑3 static analysis tools just for typechecking” and inconsistent error messages.
  • Consolidating the workflow would directly relieve the fragmentation complaints and increase confidence in LLM‑produced Python.

Haskell‑LLM Playground & Documentation Hub

Summary

  • An online REPL and snippet library that hosts vetted, fully‑typed Haskell examples ready for LLM copy‑pasting, with instant type‑checking and documentation.
  • Lowers the barrier to using Haskell with LLMs by providing ready‑made, tested building blocks.

Details

Key Value
Target Audience Researchers and engineers experimenting with LLM‑generated Haskell, especially those evaluating Haskell for production use.
Core Feature Hosted sandboxed REPL, searchable library of copy‑pasteable Haskell snippets with verified type signatures, and CI‑validated documentation pages.
Tech Stack React front‑end, Node.js back‑end, GHCjs for safe execution, GitHub Actions for CI verification of snippets.
Difficulty High
Monetization Revenue-ready: Pay‑per‑usage API credits + enterprise subscription.

Notes

  • Commenters ask why not use a fast‑compiling language; a curated Haskell snippet hub would give them the safety of strong types without the compile‑time penalty.
  • Providing reliable, pre‑checked examples directly addresses the need for trustworthy LLM‑generated Haskell code.

Read Later