IdeaWell

Project ideas from Hacker News discussions.

Kev: Tiny Jev-like family of decision models built on top of Qwen3.5

Original Article

Hacker News Discussion

📝 Discussion Summary (Click to expand)

Theme 1: Practical use cases for fast decision models
Users highlight scenarios where a lightweight classifier can replace heavier LLMs or manual logic—such as routing, verification, or simple “smart if‑statement” checks.
- “Been hoping for something in this space. Jev‑like decision models on Qwen3.5 could really simplify some of our internal routing logic.” – dunlin
- “You should call Jev‑like models when you give it a JSON‑like structure to produce, it is useful when you need some intelligence in your code. … I want to add that you can see Jev like a smart if‑statement.” – lucrbvi
- “The most obvious usage would be to have local fast cheap 'feedback' / 'control' over a slower more expensive agent … Things like 'goals' could now be split from a long prompt into 'actions' and 'verifiers'.” – NitpickLawyer
- “To develop a smart ai system for my 2d roguelike platformer? … its low latency entices me.” – saejox
- “Consider every situation where you 'force' an LLM to output only a choice / category, or a set of them. If you have workflows like that, you're now being promised significant cost‑ and latency reduction.” – vidarh

Theme 2: Latency and cost advantages
Many commenters stress that Jev‑style models run much faster and cheaper than full‑scale LLMs, enabling real‑time or low‑budget applications.
- “significant cost‑ and latency reduction” – vidarh
- “its low latency entices me.” – saejox
- “On Gemma 4 12B, I am getting 220 ms per move … So, if a 12B dense model can offer this latency on a local old PC, then definitely you can scale it up …” – faangguyindia
- “local fast cheap 'feedback' / 'control' over a slower more expensive agent” – NitpickLawyer

Theme 3: Debate over novelty and comparison to existing classifiers
A recurring thread questions whether Jev offers anything truly new or merely repackages known classification techniques, sparking both enthusiasm and skepticism.
- “I don’t think this is a very good use case. You can already do this much better with a normal LLM.” – jeeeb
- “Bit of a Jev explosion going on. Is it because it's taking us back to a simpler time we understand better? Classification models have been around for a while.” – monkeydust
- “Jev is creating a sort of identity crisis for me, because the number of absolutely clueless folks parroting the classifier thing is the first time I've seen this sort of mass psychosis in CS upfront.” – BoorishBears
- “I think that there are not that many use‑cases that have been opened up by this that tool‑calling on other models didn't solve already.” – petesergeant

🚀 Project Ideas

Generating project ideas…

JevVerifier

Summary

  • Provides a fast, low‑cost verification API that uses Jev‑like decision models to validate LLM‑generated code actions against project conventions.
  • Core value: replace expensive LLM verifiers with sub‑100 ms, confidence‑scored checks, cutting agent loop latency and cost.

Details

Key Value
Target Audience Developers building LLM‑based coding agents or AI‑assisted IDEs
Core Feature Action‑verification endpoint that takes a JSON plan item and returns pass/fail + confidence
Tech Stack Python (FastAPI), ONNX runtime, distilled Jev model (Qwen3.5‑based), Docker (optional K8s)
Difficulty Medium
Monetization Revenue-ready: Usage‑based pricing ($0.001 per 1k verifications)

Notes

  • HN users emphasized the need for cheap “feedback / control” over slower agents: “You could have local fast cheap 'feedback' / 'control' over a slower more expensive agent … verifiers … decide if the step was done correctly” (NitpickLawyer).
  • The idea of using Jev‑like models for JSON‑like verification was highlighted: “You should call Jev-like models when you give it a JSON-like structure to produce … useful when you need some intelligence in your code” (lucrbvi).

StyleGuard

Summary

  • VS Code extension that runs a local Jev‑like model to classify UI/component code snippets as “on‑brand” or “off‑brand” and suggests fixes.
  • Core value: real‑time design‑system compliance without manual reviews or heavy linting pipelines.

Details

Key Value
Target Audience Frontend engineers, design system maintainers
Core Feature Real‑time classification of JSX/TSX/CSS snippets against a configurable design‑system rule set
Tech Stack TypeScript VS Code extension, ONNX/WebAssembly Jev model, JSON rule config
Difficulty Medium
Monetization Revenue-ready: Per‑seat SaaS ($5/user/month)

Notes

  • Commenters noted the pain of manual style enforcement: “I think a great use case for these will be when they have large context windows and are able to enforce styling rules for frontend development … you can't run into drift issues” (nullbio).
  • The ability to run locally without heavy tooling aligns with the desire for lightweight, fast decision models: “It’s fast enough and cheap enough that you can hook it up to a game loop … real time” (Fabricio20).

SpamSniper

Summary

  • Self‑hostable microservice that classifies incoming text (email, comments, form submissions) as spam or ham using a Jev‑like model, returning a confidence score.
  • Core value: ultra‑low latency (<50 ms) spam filtering at a fraction of the cost of large LLM‑based classifiers.

Details

Key Value
Target Audience Indie SaaS founders, forum/mod admins, email service providers
Core Feature Low‑latency spam/ham classification API with confidence scores
Tech Stack Go HTTP server, TinyML Jev model (flatbuffers), Redis for caching, Docker
Difficulty Low
Monetization Revenue-ready: Hosted API tiered ($0.0005 per 1k requests)

Notes

  • Users pointed out the high cost of LLMs for spam: “I have been using 27B‑class models for a while … they are nearly perfect at determining what is spam … disadvantage is computational cost” (jwr).
  • The appeal of sub‑400 ms responses for real‑time use cases was highlighted: “It has a reply in less than 400ms … you can hook it up to a game loop … highly state dependent” (Fabricio20).

Read Later