There’s a lot of AI hype out there — flashy demos, powerfull AI-Agents, auto-generated pitches, ChatGPT wrappers everywhere... But I want to hear ...
For further actions, you may consider blocking this person and/or reporting abuse
Ali Farhat on July 04, 2025
For further actions, you may consider blocking this person and/or reporting abuse
Thanks for sharing — this really hits home. Your tool looks like a great way to get actionable insights quickly, and I think it might be helpful when I make suggestions to my clients.
I’m building Kaizen Agent, an open-source AI QA tool that tests and auto-fixes LLM apps. I got tired of the cycle: write → test → fix prompts and code → test again. It was slow and boring, so I built a tool to automate it — and now I use it for my own agents.
GitHub: github.com/Kaizen-agent/kaizen-agent
Thanks @suzuki_yuto_786e3bc445acb
Just out of curiosity, what do you mean exactly with fixing prompts? Does it need to meet a certain criteria?
Fixing prompts means tweaking them when the output from your LLM or agent doesn’t meet your expectations. The criteria can vary depending on the use case, but the goal is to improve the prompts (or code) so your LLM app can handle a wide range of user requests reliably.
That’s exactly what we’re trying to help with — making it easier to test, identify failures, and iterate quickly.
Great post, Ali! Loved reading about your AI tool—it’s always inspiring to see real, hands-on projects rather than just theory. It's awesome how you opened the floor for others to share too. The community needs more of this kind of practical exchange!
Thank you! 🙏 I would love to see other projects as well 🙂
Great question! I just released Voice AI Workforce - a React component library that solves a problem I was constantly running into: building voice interfaces for different user types.
The cool part? It's the same voice interface that automatically adapts based on who's using it:
Same voice command → completely different UI/responses depending on the user type. No more building 3 separate interfaces! 🎯
It supports multiple AI providers (OpenAI, Anthropic, Google) and is fully TypeScript with React components that just work.
GitHub: devvenueboost/voice-ai-workforce
npm: @voice-ai-workforce/react
The best part? It's open source and I'm looking for contributors to help add more AI providers and build Vue/Svelte versions.
What's the most frustrating part about building AI interfaces you've encountered? Always curious to hear what problems other devs are facing! 🤔
This looks sick brother! Also already 600+ npm downloads! Did you ever heard of Vapi? This is also an AI Voice Support for inbound calling. The only downside that it is US based only, and I am from Europe.
Do you happen te know any cheap Voice AI agents for outbound calling?
Thanks bro
Yes i am aware of vapi ai
Hmm not sure about any cheap voice ai agents for outbound calling 🤔
I built a moderation bot that hides scam replies on my company's Twitter. We're in crypto, so unfortunately, scammers are constantly trying to victimize our users. The bot uses the Twitter Filtered Stream API and sends new Tweet replies to ChatGPT 4o to detect if they are scams. If they are scams, it hides the reply. It's very effective with low false-positive and false-negative rates. It's actually quite simple and not impressive at all from an engineering standpoint, but I'm proud that it protects users and I would guesstimate it's prevented over a million dollars in losses many of which could have been devastating for the victim.
Amazing that it prevented that many losses 🙌 What is the current false-positive rate?
It's hard to say exactly but most false-positives are of a specific type: they're replies that contain just a link to a website, and no other text. The bot can't follow the link (yet) so it has to just look at the URL and decide if it might go to a scam website. It does that about as well as can be reasonably expected, but it does hide a fair few non-scam links. Fortunately we can easily un-hide them when we notice that.
For tweets that actually contain text, the false positive rate is pretty close to zero, I would estimate maybe 1 in 1000.
Really cool tool! Love how practical and fast it is. We’re building something similar for internal HR automation using Slack + Make.com + GPT — already saving our team a lot of time. Thanks for sharing!
Thank you! Ah yes, we've built several HR tools using AI
I made jonah.mileswk.com. This is a piggy AI Discord bot that talks back to the humans!
This is very interesting! How many users add the pig to the server?
Not many 😭. I can't find a way to get it out there.
Try placing it on Product Hunt, you will have my vote 🚀
Btw, how do you monetize this? I guess when lots of people using it you can add some advertisment in messages 🙂
We recently built an internal HR AI agent that helps new employees onboard faster. It answers policy-related questions, helps book training sessions, and even schedules intro meetings with team leads — all integrated with Google Calendar and Slack.
Was surprisingly easy to set up with a Node.js backend + GPT + Make.com handling most of the workflows. Still testing, but already reduced HR team load by around 30%.
That sounds awesome, especially the calendar and Slack integration. We’ve been thinking about expanding ours with internal HR flows too.
Right now we’re mainly focused on external-facing agents like lead intake, qualification and onboarding, but internal ops is definitely on the roadmap. Curious how you handle fallback logic when the agent doesn’t know the answer. Do you route to a human or just log it for training?
I had a similar idea like this, but never got the change to develop it, but this gives me more ideas for my next project. :)
Btw, which GPT you use and what is there average cost per scan?
About the GPT and costs, it really comes down to how much data you feed the model (token usage). I initially used GPT-4-turbo for the first version, but switched to GPT-3.5-turbo since it’s about 10x cheaper, and for this type of structured scan the difference was minimal.
As a result, the average cost per scan is well below $0.01 — not even a cent. That makes it super scalable, especially when combined with caching or batch logic.
Let me know if you ever pick it up again, happy to exchange notes!
Looks great! But I would focus more on AI in general, now it also outputs non-related AI solutions
Thank you :)
Yes, we are still looking for that sweet spot to offer all services combined with AI
I guess people who are using the tool would use it to discover AI solutions, but I can imagine that you want to offer all your stuff, anyways, beautiful tool :)
🙌 🙌 🙌
It was interesting to read about your AI tool.
I also appreciate your openness in inviting others to share their work.
Practical exchanges like this are valuable for the community.
For sure, thank you 🙌
I am looking for someone who can help us built an AI sales againt
Sure, we can help you with this :)
Very nice!
New update! The tool should give a better response now
This is an interesting tool, you will see a lot more of these in the near future
🙌 🙌 🙌
Great work brother!
Thanks! 🙂
Let's go AI people :D
We really need to get our hands dirty as well as it already feels that we are behind
Yes, doing nothing will become very expensive