What are you building? Drop your AI product below
by•
Doing one of these because I get more from reading what others are building than from any newsletter.
I'll start.
Building AI Hive, an enterprise AI agent platform that helps mid-market and enterprise teams get from AI pilot to production in weeks instead of quarters. The hard problem we keep solving: compliance, model flexibility, and the lack of in-house AI talent on the customer side.
Product Hunt page: https://www.producthunt.com/products/ai-hive
Your turn. Drop:
- What you're building
- Who it's for
- The hardest part you're solving right now
Will read everything and upvote what resonates.
2.4K views
Replies
siOsi – upload one selfie and it reads your 12-season color type, undertone, gold vs silver, and which hair colors suit you, then grades the actual makeup in the photo (flash cast, oxidation, coverage), which most tools skip. Free, no signup. Built it because every beauty sub has a queue of "what season am I" posts waiting days for three strangers to give three different answers.
@aktivist This is such a clever niche Demetrio. Every beauty sub really does have those endless "what season am I" threads with three conflicting answers, so solving it with one selfie is smart. The fact that you grade actual makeup application (flash cast, oxidation) instead of just color type is the part most tools skip, nice touch.
@nolan_vu , thanks – and that grading part is what nearly killed the project. Judging applied makeup means separating the camera from the face first: flash cast and white balance move the pixels more than the foundation does, so oxidation only becomes measurable after you normalize the light. That's exactly why most tools stop at color type – a color read survives bad lighting, an application read doesn't. (We go live on PH Aug 22, so this thread's been good practice at saying it in one paragraph.)
Love this format, Nolan, thanks for kicking it off.
What I'm building: IntelliResearch, an all in one AI research assistant. You type a topic once and a dozen agents run in parallel: papers across six databases, then a literature review synthesized only from those papers (every claim cited, not hallucinated), plus a comparison table, research gaps, datasets, code, patents, conferences, and grants.
Who it's for: researchers drowning in tabs. PhD students, postdocs, faculty, and lab teams who currently stitch together eight separate tools for one literature review.
Hardest part right now: trust. Researchers will only stake their work on output that's grounded in real papers and cited back, with zero hallucinated references. Getting that while orchestrating a dozen flaky scholarly APIs (different schemas, rate limits, downtime) into one fast response is the whole game. It has to feel instant and never make anything up.
https://intelliresearch.tech
Checking out AI Hive now. The pilot to production gap is very real.
@sajid_javid The "every claim cited, not hallucinated" part is doing a lot of work here and it should be front and center in your positioning. Researchers I've talked to don't hate AI tools, they just can't stake their name on outputs they can't verify. That trust angle is your whole moat.
@nolan_vu Thanks, Nolan. That's a really helpful perspective. I originally focused on the breadth of features, but you are right, the real problem isn't access to AI, it's confidence in the output. We are leaning heavily into grounding every insight in real papers with verifiable citations, and I will definitely revisit our positioning to make that the first thing people see.
@sajid_javid thanks for your reply and wish you best of luck with the product
What I'm building: Consile, an all-in-one people operations platform, HR, payroll, tasks, attendance, and recruitment, unified in one workspace instead of five different tools.
Who it's for: growing teams (5-100 people) who've outgrown spreadsheets but aren't ready for expensive enterprise HR suites.
The hardest part I'm solving right now: distribution, more than product. The build itself has gone well, real security, real users, recently shipped native MCP integration so it works directly inside Claude and ChatGPT. But going from "solid product" to "enough people know it exists" is proving to be a genuinely different problem than building it was.
Product Hunt page: https://www.producthunt.com/products/consile
@njoroge_emmanuel "Solid product to enough people know it exists" is such an accurate description of where most good B2B tools get stuck. The MCP integration with Claude and ChatGPT is a smart move though - that's distribution that doesn't feel like distribution.
Upvoted and following Consile. Would love it if you had a moment to check out AI Hive - we're also in the B2B automation space and your feedback would mean a lot: https://www.producthunt.com/products/ai-hive
@nolan_vu Appreciate that, genuinely, and good instinct calling out the MCP angle, that's exactly the framing I've been leaning into, distribution that comes from being genuinely useful inside a tool people already have open, rather than another channel to manage.
Checking out AI Hive now, happy to give real feedback, not just an upvote.
I am currently building my first product: Solstice. It is an all-in-one SaaS platform for highly customizable AI solutions. Right now, I am implementing a seamless self-serve onboarding flow: customers simply fill out their specific requirements and instantly receive a fully personalized AI chatbot ready to be embedded directly into their website. During this final onboarding stage, different customer segments simply select the tools they want to activate, inject their custom context, and the chatbot is instantly ready to deploy. I am in early stage. Built the RAG pipeline, the website frontpage and the onboaring page. Currently iterating on personalization.
@alelusardi Thanks for sharing this, the self-serve onboarding angle is something I find genuinely interesting. Getting customers to a personalized chatbot ready to embed without needing a sales call is a real UX unlock if you can pull it off cleanly.
Would love to see how it goes once the personalization layer is live
What I'm building: AdWait, a Chrome extension that shows one small labeled ad while your AI is generating and credits you half of what it pays. It disappears when the response lands. No account, and it never reads your prompts.
Who it's for: people who use AI tools all day, and specifically agentic ones. A ten-minute Claude or Replit run is about 85 ad rotations of a developer watching a progress indicator with nothing to do.
Hardest part right now: finding advertisers. I can build the thing. I have no idea how to sell it, and it's two-sided, so users with no ads to show them is a dead product. Cold outreach as a nobody is its own specific problem. I'm emailing companies with no case study, no audience number worth quoting, and a launch date where a track record should be. The honest pitch is "this is cheap and it's a bet," which is a real offer to exactly one kind of buyer and a waste of everyone else's time. Finding that buyer faster is the entire problem I'm working on.
Launch is planned for next month and I have zero advertisers. Users with no ads to show is a dead product, so the demand side is the entire game for me right now, and it's the part I'm worst at.
https://www.producthunt.com/products/adwait?launch=adwait
If you sell to developers, my inventory problem might be your cheap channel: ads@adwait.io
@sjtkang The two-sided marketplace cold start is genuinely one of the hardest things to solve without a track record to show either side. The honest pitch of "it's cheap and it's a bet" actually respects the buyer's intelligence more than most founders admit, so I think you're closer to the right framing than you realize.
Hope you find that first advertiser soon, that's usually the unlock that makes everything else easier to sell. Upvoted and followed! If you're open to it, would really appreciate a look at AI Hive too, it's a workflow automation platform for teams trying to actually ship AI into production: https://www.producthunt.com/products/ai-hive
I'm working on something I’m really excited about: ArchiKist! It’s a free toolbox packed with calculators and practical tools designed for architects, builders, students, designers, and even DIY homeowners.
Basically, it’s for anyone who’s ever found themselves asking, “How much do I need?” or “How do I figure this out?” without wanting to dive into spreadsheets or sift through a bunch of different websites. 😅
Right now, I’ve included tools for things like paint, tile, roof pitch calculations, FAR, occupant load, scale, unit conversions, room calculations, and so on. I’m adding more features gradually as I go along.
The biggest challenge has been making sure these tools are really user-friendly while still nailing the calculations accurately. It’s amazing how easy AI makes it look to create a polished product, but there are always tricky edge cases and little UX issues that pop up.
I launched about 16 days ago, so I’m still in that stage of building, watching how people use it, fixing things, and then building some more. I’d love to hear what everyone else is working on too! There are always some really interesting and unique projects in these discussions.
@gala_app Thanks for sharing, Gala. ArchiKist sounds like one of those products where accuracy matters way more than flashy features. The challenge of balancing usability with trustworthy calculations is very real. Excited to see how the toolbox evolves as more people start using it in real projects.
@brajn
The decay mechanic is genuinely interesting. Most attention platforms fight decay; you're building the decay in as a feature. One thing I'd push on though: in practice, does transparent math actually feel more fair to users, or does it just shift the advantage to whoever has more refueling budget?
I ask because we deal with something adjacent in enterprise AI — "explainability" is a selling point until someone realizes the explanation is still complex enough that only engineers understand it. Curious if you're seeing a gap between users who say they want transparency and users who actually engage with the ranking mechanics. Also, the £10/€10/$10 starting credit is a smart way to remove first-move anxiety. What's the conversion looking like from credit-holder to active Slot owner?
@nolan_vu
Excellent questions Nolan. Regarding the first point, the transparent math definitely shifts the dynamic, but we balance the refueling budget by capping how much energy a slot can hold at any single moment. This prevents a massive corporation from buying a top position for a year in advance. Because the decay rate is constant and universal, a smaller business with active community backing can outmaneuver a larger budget through timely engagement and organic boosts from supporters. The math is kept simple, just basic addition and subtraction of energy over time, so you do not need an engineering degree to understand where you stand.
About the user gap, we do notice that casual visitors just care if the board looks alive, while power users and creators immediately dig into the mechanics to optimize their visibility windows.
As for conversion, the starting credit completely changes the onboarding behavior. Instead of staring at an empty dashboard, new users immediately claim a slot to test the water. Currently, around forty percent of credit holders go on to set up a complete slot profile with images and active links, while the rest mostly use their tokens to experiment with boosting existing creators on the leaderboard. It is a fascinating micro economy to observe during this testing phase.
@brajn what a quick response, totally agree with your sharing points
@brajn are you the maker or hunter of any product related to AI, since I am in that niche so I think we can help each other in providing feedback or recommendations
What we're building:
An AI-powered options trading companion for Indian retail traders. Finn analyzes Nifty, Bank Nifty, and selected stock options to surface actionable trade opportunities with recommended entry ranges, target levels, risk parameters, and expected holding periods. The goal isn't more data—it's better decisions.
Who it's for:
Retail options traders who are serious about improving their edge but don't have the time, tools, or institutional resources to analyze every opportunity. People who want structured trade ideas and decision support instead of market noise and endless charts.
The hardest part we're solving right now:
Trust. In trading, a recommendation is only valuable if a trader has the conviction to act on it. We're focused on making every trade idea transparent, explainable, and easy to evaluate so users understand not just what the opportunity is, but why it exists.
The future of retail trading isn't fully automated execution—it's AI-assisted decision-making with the trader firmly in control.
We're currently in beta, working closely with early users, collecting feedback, and refining the experience with every iteration.
Would love to connect with builders working in fintech, investing, or AI-powered decision support. 👇
@daksh_jain16 The trust framing is dead on. In trading, even the best recommendation is useless if the user can't understand the reasoning behind it. Most AI trading tools fail at exactly this point, they spit out signals without explaining why, and serious traders just won't act on a black box no matter how accurate it claims to be.
The "AI-assisted decision making with the trader firmly in control" positioning feels like the right path for retail, especially in Indian markets where regulation around fully automated execution is still tricky. How are you measuring whether users actually trust the recommendations enough to act on them? That'd be a fascinating metric to watch through the beta.
Building ReplyReview, an AI tool that helps mobile app developers manage and reply to App Store and Google Play reviews faster.
It's built for indie developers and small studios that don't have time to manually answer hundreds of reviews.
The hardest problem I'm solving right now is turning raw user feedback into actionable insights instead of just generating AI replies. I'm focusing a lot on daily summaries, issue detection, and helping developers spot trends before ratings drop.
Would love to hear how others are handling app reviews and user feedback.
@frangimenez_dev thanks for dropping this, Fran!
the shift from "generate AI replies" to "surface actionable insights before ratings drop" is the right pivot and tbh the harder product to build. most tools stop at the reply layer because it's the visible output, but the early warning signal on trend detection is where the real retention value lives. would love to know how you're handling apps with reviews in multiple languages when doing issue detection.
Good product.
We are building ServBay, an AI-native local development base that simplifies web dev and empowers AI coding agents.
What we're building: ServBay. It’s a lightweight, native alternative to Docker on macOS that bundles 50+ dev services (databases, languages, web servers) with one-click setup, alongside a local-first AI Gateway and built-in MCP server.
Who it's for: Indie hackers, solo developers, and dev teams who want to move fast, save system resources, and seamlessly integrate AI coding agents (like Claude Code or Cursor) into their local development flow.
The hardest part we're solving right now: Security and local-cloud orchestration. Specifically, how to keep LLM API keys 100% local (zero-leak) while building a secure, sandboxed bridge (via MCP) that allows AI agents to inspect local databases, configure services, and read logs without messing up the developer’s host machine.
Website: https://www.servbay.com
Product Hunt page: https://www.producthunt.com/products/servbay
@saltyfishh thanks for this breakdown, the local-first angle is underrated.
the zero-leak API key problem is actually one of the most annoying unsolved things in local AI dev right now. most setups require you to either trust the cloud or cobble together your own proxy, so a native sandboxed MCP bridge that keeps keys local while still letting agents inspect databases sounds like the right architecture. curious how you're handling key rotation when the agent needs to re-authenticate mid-session.