PH热榜 | 2026-04-19

Google Research just made the hardest skills to measure, actually measurable.

Vantage is a Google Research experiment that uses GenAI to assess future-ready skills like collaboration, critical thinking, and creativity. AI avatars simulate real scenarios, score your performance, and deliver a personal Skill Map.

The problem: Critical thinking, collaboration, and creativity matter most but are nearly impossible to assess at scale.

The solution: Vantage uses an Executive LLM to simulate real team scenarios, surface skill evidence, and score performance at human-expert level.

What stands out:
🧠 AI simulated team: Work through missions like debates, pitches, and experiments with AI avatars.
🎯 Executive LLM: Introduces dynamic challenges like conflict and constraints mid-conversation.
📊 AI Evaluator: Scores using expert-level rubrics with human-like agreement.
🗺️ Personal Skill Map: Visual scores with precise qualitative feedback.
🔬 Validated by New York University: AI scoring matches human experts across 188 testers.
📐 Aligned with OECD and World Economic Forum frameworks.
🎓 Built for classrooms: Designed as a skills layer alongside existing curricula.

Skills assessed:
- Collaboration: Conflict resolution and project management.
- Creative Thinking: Generating, building, and evaluating ideas.
- Critical Thinking: Interpreting, analyzing, and judging information.

Different because it’s not a test, but an adaptive conversation that reveals real capability.

P.S. I hunt the latest and greatest launches in tech, SaaS and AI, follow to be notified → @rohanrecommends

Interesting idea! How do you ensure the AI-simulated scenarios stay relevant as tech stacks and best practices evolve? I remember one painful incident where outdated training led to a junior engineer deploying a container with root privileges to production.

This would be a great brainstorming partner!

critical thinking and collaboration are the skills I need most but never know how to actually benchmark. curious how the skill map tracks improvement over time. do repeated sessions show progression?

Cool! Built something like this before under the guise of an interview and its interesting how a reframing from interview to assessment enables you to discover a whole value add of education.

什么时候有Windows版本？

Hi Product Hunt 👋 I’m Adrian, cofounder of Verdent.

With most AI coding tools, you are still acting as the PM, engineer, QA, and ops. You keep switching contexts, checking every step, and re explaining things every time. Instead of building a business, you end up managing the process. Most tools help you prototype. Verdent is built to help you actually ship.

Stop wearing every hat

Say “I need a booking page with time slots and payments.” Manager handles the planning, implementation, and testing. You just decide what to build next.

Stop managing every step

Give it a goal and come back to something that works.

Stop repeating yourself

Manager keeps track of your project, your standards, and how you like things done. It gets better the more you use it.

Stop being tied to your desk

Send it a message from your phone. It keeps working while you are away.

Our most active users are solo founders shipping features daily without writing code.

Last launch was about coding with AI.

This one is about helping builders actually run and ship their business.

Try it free 🙌

Coolest launch of the day but "AI technical cofounder” is a strong claim. What responsibilities do you think should never be delegated to Verdent?

Hey all, we have integrated Anthropic's latest model Claude Opus 4.7 already! Excited for everyone to try it in Verdent：）

Looks exciting, couple questions though. When Verdent is "working even when you're offline", do you mean it is running agents on my infrastructure, or cloud-side on Verdent's? Curious how you handle context limits across long async runs.

Does it maintain shared context across agents or does each run independently?

Personal Computer by Perplexity AI is redefining what a computer can do.

Most computers still rely on you to juggle files, apps, tabs, and workflows. Personal Computer flips that, it acts as an AI orchestrator that understands your objective and executes tasks across your local files, apps, connectors, and the web.

What makes it different? It blends local + cloud environments, bringing multi-model orchestration directly to your machine, not just inside a chat window.

Key features:

• Works across files, native apps, and the web

• Automates complex & continuous workflows

• Reads and executes your to-do lists

• Organizes files and compares local + web data

• Voice + keyboard activation (CMD shortcut)

• Secure sandbox with auditable, reversible actions

Who it’s for & use cases:

• Builders, operators, and power users

• Automating repetitive workflows

• Managing files, tasks, and communication

• Decision-making using local + web context

Why it matters: This is a shift from “using a computer” to managing an AI that uses the computer for you.

If you’re into the future of AI-native workflows, this is worth watching.

P.S. I hunt the latest and greatest launches in tech, SaaS and AI, follow to be notified → @rohanrecommends

The interesting shift here isn’t voice or integrations, it’s state. If the system can remember what it was doing across files, apps, and sessions, that’s when it starts to feel like real work is moving. Stateless AI is helpful, but it doesn’t reduce ownership. How persistent is the system across sessions?

If this is essentially an alternative to Claude Cowork - impressive how quickly you shipped it. The "agent that actually does stuff on your machine" space is heating up fast.

Hey Product Hunt 👋

We’re Vivek and Mike, co-founders of Avina.

We built Avina because the bar in go-to-market keeps rising, but tools haven't kept up. Buyers ignore generic emails and poorly targeted ads, yet most prospecting and automation platforms still hand you broad lead lists and noisy "signals." Running GTM intelligently means duct-taping tools and complex workflows together yourself.

Avina does it for you. The best GTM starts when buyers actually have a reason to care, so Avina finds the right people at the right moment and helps you get in front of them wherever they are.

How it works:

1. Describe the kinds of leads that need your product and the triggers that suggest they're ready to buy, all in plain English.

2. Avina finds those leads, enriches them with verified contacts, scores them against your ICP, and refreshes your audience daily.

3. Then it gets you in front of them across personalized AI emails, hyper-targeted ad audiences, and ABM campaigns.

4. Everything syncs back to your CRM and Slack.

A few examples you can try:

• "Healthcare companies that had a data breach in the last 6 months"

• "Prospects who spent 30+ seconds on our pricing page in the last 2 weeks"

• "Insurance carriers that acquired a competitor in the last 90 days"

• "Companies whose filings mention efficiency initiatives and are hiring ops roles"

Website de-anonymization (RB2B, Vector, Clearbit, and more) is included out of the box, so you can act on first-party intent too.

The goal:

Go from signal to warm lead to relevant outreach across every channel your buyers live on, without duct-taping ten tools together.

PH-exclusive deal:
15% off for six months

We’ll be in the comments all day and are excited to hear what you think!

Congrats on the launch! Awesome team!!

Congrats Vivek and Mike! I’ve used Avina at both of my last two companies and it provides such fantastic intelligence and customer buying signals.

Keep it up team!

Awesome job Mike and Vivek!

As a former Product Marketer, selling NetOps tools to SecOps and vice versa, I wish we had something like Avina to help capture exactly when a network buyer persona is looking to expand into security services. Or when an IT org is going through a re org and the CISO goes from reporting to the CIO straight to the CEO.

This sort of intelligent GTM agent would have made my life a lot easier when I was a marketer. I can see a lot of MSSPs get use from this as well. It would be cool to see Win/Loss insights as well that can inform product strategy. Great stuff and congrats on the launch!

Amazing! Finally we can close our Apollo subscription. For the prospectives, are they searched online in realtime, or you have a proprietary database?

Hey Avina Team,

For me as a marketer, the most important thing is understanding exactly which signals the tool can use as triggers, beyond generic and well-known signals like a company raising funding.

BR,
Oleksii

Congrats on the launch, team! What was the motivation to build this app? Personal experience or a gap that you spotted in the market? What would you say your biggest differentiator is against other agentic AI GTM tools that are available?

Congrats on the launch! This looks like a solid all-in-one solution for creators juggling multiple platforms. I'm curious about the AI thumbnail generation - how does it handle brand consistency across a creator's channel? Does it learn from past thumbnails that performed well, or is it more of a one-off generation tool? Would love to hear how creators are using it in practice.

feeding a reference URL to extract visual DNA instead of describing a style from scratch. that's the part I've been doing manually for too long.

The problem is real. Comment windows on Instagram are tiny and the algorithm rewards early engagement heavily. Does this work across multiple accounts or is it tied to one?

Why did you decide on YT, Twitch and esp Instagram? I think that YT and Twitch are the same category, but whatIG?

The native Instagram notification feed becomes impossible to manage once a reel gets any real traction. I wonder if you are relying on the official Graph API to pull these in real-time or if you had to build a custom scraper. An automated filter that specifically surfaces comments containing questions would make this a total no-brainer.

Hehehe love this kind of widgets that makes our work funnier! It's a matter of enjoying the journey as well. Wish you all the best Will

Congrats on the launch! This is such a fresh take on system monitoring - the 3D object approach is genuinely delightful. I'm curious about performance impact: since these are animated visualizations running on the desktop, how optimized is Tell to avoid becoming a resource hog itself? Do you have any benchmarks on CPU/memory usage?

Does it actually delete accounts for you or just surface them? Great tool btw!

Moving the coding agent to the cloud is a smart play since running large context loops locally always sets my laptop fan on fire. I am really curious how Fixa manages secure access to existing private repos to understand an established codebase. Letting this loose on a backlog of minor UI bug tickets would be a perfect way to test its actual reasoning limits.

Congrats on the launch! This is genuinely impressive. Quick question - when Fixa autonomously installs dependencies and writes backend code, how do you handle security validation? Like, are there guardrails to prevent it from installing malicious packages or introducing vulnerabilities, or does it require human review before deployment?

Hey Product Hunt, I'm Jeremy, and I built Wyndo.

Here's the problem I kept having: I'd check the weather before a run, see "60% chance of rain," then squint at the hourly graph trying to figure out when in the next three hours that 60% actually lives. I'd re-open the app three more times before heading out. Sometimes I'd leave too early. Sometimes I'd get caught in it anyway.

Wyndo is the weather app I wanted. Pick your activity — walk, run, bike, drive, or eat outside — and it scores the next few hours and tells you go, wait, or skip it, with the reasoning right there on screen. Every recommendation cites the weather factors you can already see: precipitation timing, wind, temperature, visibility, alerts. A light drizzle that's fine for a drive gets flagged for outdoor dining.

The engine is deterministic and rule-based. No AI guessing at the forecast. LLMs help parse questions and phrase answers. Data comes from OpenWeatherMap for minute-by-minute precipitation, Open-Meteo for hourly and daily forecasts, and the National Weather Service for alerts. Free, no ads.

Try it at wyndo.app — no signup needed to ask your first question. I'd love to hear what you ask it, and especially where the answer felt wrong or thin. The feedback form is one click from anywhere in the app, and responses feed back into the eval for future recommendations.

Simple but cool product! Btw you explicitly avoided ai for decision making, so do you see a future where learned preferences (using ml or something) outperform deterministic rules here??

Big congrats on the launch! 🚀 I love the 'go, wait, or avoid' approach. I’m constantly looking at the 40% chance of rain and trying to do the math in my head on whether I’ll get soaked if I take the dog out now or in an hour.

Taking that mental work out of the equation is such a smart idea.

Hey Product Hunt,

I’m Rénald, founder of Cohesium AI.

I built Assemble because I was tired of AI tools that sound helpful but stay generic. A code review becomes a polite summary. A security audit becomes a reformatted checklist. A multi-step project starts strong, then falls apart as soon as context gets longer or the work gets more complex.

So I built what I actually needed: a structured AI work system, not just another assistant.

With Assemble, you type /go and describe what you need. From there, it routes the task by difficulty, keeps useful cross-session memory, and switches into a spec-driven workflow when the work is complex. For bigger delivery, it can even move execution into a board with review and test stages.

What makes it different from most agent frameworks:

• it’s a configuration generator, not a runtime
• zero daemon, zero SDK, zero dependencies, zero lock-in
• native configs for 21 platforms including Cursor, Claude Code, Codex, Gemini CLI, Copilot, and Windsurf
• it works beyond coding too: docs, contracts, proposals, email, and client operations

The Marvel framework isn’t branding — it’s a prompt-engineering choice. In testing, it gave us stronger role identity, better consistency, and less generic output than traditional agent setups.

And because LLMs naturally agree too easily, Assemble bakes in structural dissent: Deadpool challenges assumptions by default, and Doctor Doom escalates high-stakes decisions.

A real turning point for me: a client project that was supposed to take 2 days turned into 10 days of failed attempts with generic AI tools. With Assemble, it took 30 minutes.

If you try it, I’d genuinely love your feedback — especially on the workflows, platforms, and specialist roles you’d want next.

MIT licensed. Open source. Built for real work.

Congrats on the launch! This looks really impressive. I'm curious about the memory component - when you say it "remembers," does that persist across different AI platforms automatically, or do users need to configure how context flows between integrations? Also, how does the zero runtime constraint work with platforms that have inherent latency?

some of the links in https://www.agglabs.com/loop are returning a 404 - i.e: https://github.com/agglabs/loop-client

Great launch Lars!

exciting one here

I’ve been using the desktop app for some time now and was eagerly waiting for the mobile version, awesome to see it finally launched! Looking forward to using it on the go 🙌

It's been a few days since I have started using the "Today" app. Unlike any other note making app, this is straightforward, minimalistic and yet has gorgeous themes. While the features like memory rail are indeed attractive, its core theme of locked notes per day truly makes for honest journaling discipline.

Been using this since beta and it genuinely feels different. No clutter, no pressure to organize, just open it and think. The midnight reset is underrated, it actually helps you move on instead of over-editing yesterday. Feels more like a mental space than an app.

The midnight lock is the reality check I didn't know I needed. I used to spend hours 'cleaning up' my notes, but Today forced me to stop performing and start actually thinking. The Memory Rail is like a conversation with my past self—it pulls up exactly what I forgot I knew, right when I need it. If you’re tired of managing a complex 'second brain' and just want your first one to work better, this is it.

Hey, long time no see 👋

We originally built Celaro in a different direction. But along the way, we kept running into the same problem:

Creating newsletters is way more fiddly than it should be.

Clunky editors. Formatting breaks. Constant workarounds.
You end up fighting the tool more than actually writing.

So we decided to focus on that.

Celaro is a code-first newsletter CMS, with a UI for content.
Think code-based building blocks, edited through a simple admin interface.

You define the structure in code. Everything else happens in a clean admin UI.

It works especially well if one of you likes working in code, and another just wants to write without touching it.

If you're building emails with React Email, SendGrid, Beehiiv, Mailchimp, or anything similar, this might be for you.

Would love to hear how you're doing it today, and what's frustrating about it.

Cheers ❤

Hey, love the idea and branding! One concern though is that this text on the landing is quite difficult to read.

Filed 5 cases against my friends and my wife already :P

We've all watched friends and family scramble through stacks of medical reports, prescriptions, and lab results during health scares - especially in India where handwritten notes are often illegible scribbles. People feel lost, anxious, and clueless about their conditions (or their kids' or parents'), needing a trusted companion to explain terms, organize files, and answer day-to-day questions without exploiting their data.

That's the problem we solved: No privacy-invasive AI "health advisor" that trains on your sensitive info or guesses at messy handwriting (we don't; we just store it safely for doctors). Instead, mediqgpt.com is your personal medical data sidekick—a secure organizer for all your docs, reports, and histories (even for pets). It explains jargon clearly, retrieves info instantly, and empowers you (and your doctor) with complete context during urgent moments - never selling or training on your data.

Our process evolved from a simple OCR tool for prescriptions to this full privacy-first intelligence layer after user tests showed the real pain: disorganized chaos in complex journeys for young kids and elders. We're not a doctor replacement, just the confident companion you need to navigate without the fear.

Try it at mediqgpt.com launching now! 🚀 #HealthTech #PrivacyFirst #MedIQGPT

I am a little worried about medical records getting uploaded on a new platform. Even though you seem not to train on the personal data but this data residing somewhere has me worried. How do you handle that?

Hey Product Hunt 👋

I work at an agency, and like a lot of you I juggle multiple projects, multiple teams, multiple GitHub repos at the same time.

The thing that was driving me nuts: having to go hunt the info down myself. Open GitHub, filter issues, figure out which PR has been sitting there for 10 days, try to understand why this bug still has no owner. And all of that without wanting to spam the team with "hey where are we on this?".

Crickets is my answer to that. I open the dashboard in the morning, I see in 30 seconds what's blocked, what's drifting, what needs me. And I close the tab. Without pinging anyone.

What it watches:

• 🐛 Unassigned bugs

• ⏳ PRs with no review

• 💬 Customer issues with no reply

• 🧟 Zombie issues (no activity for weeks)

• 🛠️ Outdated / end-of-life tech stack

• 🚌 Bus factor (one dev carrying everything)

• 📋 Repo hygiene (CI, Dependabot, security policy…)

You connect your GitHub org and that's it.

Free for 1 repo, Pro coming soon.

Really curious to hear your feedback, especially from folks juggling multiple clients or multiple teams in parallel 🙏