What Helm AI Is Building: Two Tools, One Principle

If it's your first time here and you want us in one sentence: Helm AI builds practical tools that treat AI output as a draft, never a conclusion.

This post opens a new series on the blog. We used to publish model tests only, real comparisons with public inputs and outputs. From now on the series has two halves: what we're building and why, and a product-by-product look at how things work. Anything with a test in it comes with reproducible data.

Who we are (and what we aren't)

Helm AI Inc is an early-stage software company working on practical AI tools for document work, content, and career workflows. A few facts you can check:

• Two paid products on sale: US Tech Resume Rewrite ($19 one-time) and AI Writing Assistant Pro ($4.90/month), payments via Stripe;
• We're early-stage, and our product pages say so. We don't inflate user counts, revenue, or "business readiness";
• Contact is public: the Contact page, support at support@helmaiagent.com.

What we aren't: an "all-in-one AI platform" with a pile of unrelated demos. Every product points at one specific workflow. That's the first filter we use when picking projects.

Two tools

1. US Tech Resume Rewrite: faithful before pretty

It turns Chinese (or Chinglish) work experience into English bullets that survive US tech screening. The pain it targets isn't grammar. It's quiet inflation: AI writing "Led" where you wrote "participated," adding a "40% performance improvement" your original never mentioned. Our answer is a second AI pass that checks every bullet against your source. No numbers that aren't there, no verb upgrades, no scope expansion. Try one bullet free and compare the output to your original, word by word, before you decide. There's a separate post on that checking process.

2. AI Writing Assistant: the daily writing workflow

A Chrome extension covering rewrite, tone adjustment, translation, summary, grammar checking, and reply drafting. The positioning is plain: hand the repetitive parts of daily writing to AI, but read everything yourself before it goes out. We wrote that into the product page, not as a disclaimer, but because it's our basic read on AI output: it's often wrong, just wrong in a convincing way.

One new project: Second Look

There's a project in validation worth its own introduction: Second Look, a second opinion on the advice you're getting, before you take it.

Promotion promises, family career planning, a partner's relocation, consultant pressure, a friend's investment pitch: the advice usually comes with good intentions, but the risk lands on you. Second Look breaks a piece of advice into three qualitative signals (goal alignment, autonomy risk, evidence coverage) and closes with one question: "For this advice to be good for you, what has to be true?" It doesn't judge motives and it doesn't hand out a score. It's in waitlist validation for the US market, the English page and the three-question playbook are open, and as planned it won't launch in China.

One principle: AI output is a draft, people are the endpoint

Two tools and a new project, one shared principle. On products it shows up three ways:

• Checkable: anything factual (numbers and titles in a resume) has to be verifiable line by line against the source. Our own pipeline verifies before it outputs;
• No overpromising: we don't promise interviews, we don't promise correct decisions. We promise the tool's part;
• Honest status: what's on sale, what's in validation, what doesn't exist yet. Product pages and this blog both tell it straight.

This one isn't marketing copy. We earned it the hard way: in a July test we ran, we watched mainstream models turn "participated in" into "led" on the same Chinese-written bullet. Since then, "verify, then output" has been a hard step in every product.

What this blog writes next

Three kinds, on a rolling basis:

• Project notes: what we're building, why, and what we cut (like this one);
• Product features: workflow and design trade-offs, one product at a time;
• Tests and methods: reproducible comparisons, inputs and outputs public.

If one of those matters more to you, or you want a specific product covered first, tell us at Contact. Early feedback moves the schedule.