When people hear that a shidduch organization uses "AI," they usually picture something familiar: the same kind of recommendation engine that powers Amazon or Netflix. After all, if algorithms can predict what you'll buy next or which shows someone will watch, why not who you should marry?
But these algorithms fundamentally cannot work for shidduchim.
They were built for a completely different world—a world of unlimited experimentation, low-stakes mistakes, and data so vast it can absorb millions of wrong guesses on the way to getting better.
Netflix can afford to be wrong. Shidduchim can't.
The Problem with "AI Matching"
When most organizations talk about "AI matching," they're not actually talking about AI. They're referring to machine learning techniques that date back to the early 2000s.
Many platforms are jumping on the bandwagon, marketing their systems as "AI-powered" and claiming novel new approaches. In reality, they're using the same collaborative filtering, content-based filtering, and matrix factorization methods that have existed for over two decades. Yismach tried these techniques six years ago. They performed adequately. Nothing more.
These techniques are extremely effective for retail. Amazon's recommendation engine reportedly generates 35% of their total revenue. Netflix built an entire business on suggesting what you should watch next. The technology has been refined over two decades, backed by countless research papers and billions of dollars of investment.
But these systems succeed because of two things: scale and the ability to be wrong.
Netflix runs hundreds of experiments per year, testing on hundreds of thousands of users at a time. Amazon processes billions of interactions—clicks, purchases, searches, abandoned carts—and learns from outcomes continuously. When they suggest something irrelevant, the cost is trivial. You scroll past. You don't click. The recommendation gets lost in the noise, and the system learns a tiny bit more for next time.
Shidduchim offer none of that.
A bad suggestion doesn't get scrolled past. It affects someone's dignity, their emotional energy, their trust in the process. A shadchan cannot send someone's profile to a hundred people just to "collect data." You cannot run experiments on people's lives. And when a suggestion goes wrong, the cost isn't a scroll or a skipped click—it's someone losing faith in the entire process.
The very practices that make shidduchim meaningful are precisely what starve these algorithms of the data they need to function.
How Traditional Matching Algorithms Actually Work
Beyond the experimentation, how do these algorithms actually function?
There are two main approaches.
The first is similarity-based. The system looks for people who behave like you. If Rivky and Shira both said yes to similar types of guys, and Rivky said yes to Dovid, maybe Shira would too. If Yosef and Avi both accepted suggestions from similar profiles, they probably have similar taste—so suggest to Yosef whoever Avi said yes to.
This only works when you have dense overlap—thousands of people with enough shared interactions that the system can detect meaningful patterns. In shidduchim, the data is sparse by design. Most people have only a handful of introductions. The yes rate is low. And the pool is segmented by hashkafa, community, and geography, which reduces overlap further. Rivky and Shira may have nothing in common in terms of who they've been suggested to. The model can't discover patterns that the process itself prevents it from observing.
The second is feature-based. Build a model using structured profile data—age, location, hashkafa, education, family background, learning expectations, personality traits—and predict who will say yes to whom based on which combinations of features tend to work.
This sounds more practical. But it runs into the combination problem.
The model isn't learning whether "age" matters or "location" matters in isolation. It's learning whether specific combinations of features predict a yes—this age range, with this hashkafa, in this community, with this family background, relative to this other person. The number of possible combinations grows rapidly. With just a few variables, you're already looking at hundreds of combinations. Add more features, and it explodes into thousands.
Most of those combinations will have little or no data. And in shidduchim, where 80 to 90 percent of suggestions are declined, positive outcomes are rare to begin with. The model needs enough "yes" results spread across all those combinations to learn anything meaningful—and that simply doesn't exist.
This is why traditional matching systems require tens of thousands of meaningful interactions just to stabilize. The more features you add, the more data you need. Shidduchim, by their very nature, cannot generate that volume.
And there are two more problems that make this even harder.
The first is low variance. Profiles tend to be generic—similar language, similar descriptions, similar stated preferences. When everyone describes themselves as "warm," "family-oriented," and "looking for someone with good middos," the model struggles to discern meaningful differences. There isn't enough signal to separate one profile from another.
The second is noise. The data is filled with inconsistency and mixed signals. Two people match perfectly on paper—same hashkafa, same background, same values—and one says yes while the other says no. Another pair looks mismatched on paper, yet both say yes immediately. Which signals are actually predictive? The model can't tell, because human decisions don't follow clean patterns that map neatly onto structured features.
Because of all this, traditional machine learning—so-called "AI matching"—fundamentally cannot solve this problem. We need an entirely different approach.
The Deeper Failure: Matching Words Instead of Meaning
Even if you somehow had the data, traditional systems would still miss the heart of shidduchim.
Because they read profiles like spreadsheets, not like people.
A feature-based system extracts keywords and computes overlap. A collaborative system finds patterns in who clicked on whom. Neither understands what someone is actually communicating.
But shidduchim are decided by meaning: tone, flexibility, emotional posture, expectations, what's said between the lines.
Consider this example.
His profile:
"I'm serious about learning and currently in yeshiva, but I'm also practical about the future. Looking for someone who values ruchniyus but is down to earth. I want a real person, not someone who needs everything to be perfect."
Her profile (Woman A):
"Seminary was the best year of my life. I'm looking for a ben Torah who will make learning his priority. I come from a home where my father learned for many years and I want to build something similar. I have high standards because I believe we should strive for the best in everything."
Her profile (Woman B):
"I love learning and growing, but I also know life isn't always smooth. I'm looking for someone who takes Yiddishkeit seriously but doesn't stress about every little thing. I want someone real, not a perfect image."
A traditional system sees both women mention learning, Torah, growth, values. The keyword overlap is similar. It might even score Woman A higher because she explicitly mentions "ben Torah."
But look at what's actually being communicated.
He is signaling flexibility, practicality, a rejection of rigidity and perfectionism. Woman A—despite the shared vocabulary—is signaling the opposite: high standards, idealism, a specific image of what a Torah home should look like. Woman B's language mirrors his exactly: "real," "not perfect," "doesn't stress."
The keywords match in both cases. The people don't.
This is where most "AI matching" quietly fails. It produces confident outputs from shallow understanding.
The Problem No One Has Solved: Attraction
Even if a system could perfectly match based on text—values, background, hashkafa, personality—there's another dimension that traditional matching has never been able to address: physical attraction.
This isn't something the industry ignores. It's something no one has figured out how to solve.
Attraction is deeply personal, highly subjective, and resistant to simple patterns. What one person finds attractive, another doesn't. It doesn't map neatly onto features you can put in a database. And yet it plays an enormous role in whether someone says yes or no—often before they've read a single word of the profile.
Traditional machine learning operates in the world of text and structured data. It has no way to factor in visual information meaningfully. Some have tried basic approaches—rating systems, photo scoring—but these are crude and ultimately ineffective. The problem has remained unsolved.
We researched existing methods that attempt to address this. They do not work.
So we developed our own novel model to actually understand images—specifically, attraction. To our knowledge, this capability does not exist anywhere else in the world.
What Real AI Innovation Looks Like
So what would it take to build technology that actually works for shidduchim?
It would have to function in a world of limited introductions, small data, high stakes, and dignity-first constraints. It cannot depend on endless experimentation to learn. It has to do something fundamentally different.
This is why no one has attempted it.
True AI is not machine learning rebranded. It's an entirely different capability—one that understands language the way people do, reasons through ambiguity, and grasps meaning rather than just matching keywords.
Think about what a great shadchan actually does. They don't just compare resumes. They listen. They pick up on what someone is really looking for, even when the person struggles to articulate it. They read between the lines of a conversation, notice the hesitation behind a stated preference, sense what's negotiable and what isn't. They hold dozens of people in their mind and somehow know when two of them might click.
True AI can operate in that same space. It sees through your lens—understanding preferences that are deeply personal, filtering out noise, weighing what actually matters. It takes the intuition of the best matchmakers and extends it, working on your behalf to search, evaluate, and surface suggestions that a traditional system would miss entirely.
This is not a matching algorithm that runs once and produces a score. It's an AI Match Assistant that works alongside you—learning, reasoning, explaining, and refining over time.
Building this required developing entirely new approaches and solving problems no one else has solved. It required understanding the shidduch world deeply enough to design around its constraints rather than ignore them.
No one else has done this. No one is even close.
Until now. This is what Yismach has built.
A Genuinely Novel Approach
What Yismach has built is not just AI matching. It's a system of AI capabilities that work together—each one designed to solve a specific problem that traditional approaches cannot address.
Matching beyond the written word. Most systems rely solely on what people write about themselves. But anyone who has worked in shidduchim knows that what someone writes often tells you less than what they don't write. Our system sees beyond stated preferences to detect family dynamics, hashkafic nuances, and psychological patterns that prove far more predictive of whether someone will say yes or no. Fourteen years of doing this work has taught us where to look. AI helps us see it at scale.
Visual and text working together. We don't treat attraction as a separate problem to solve later. Our system integrates visual and text-based features to provide the optimal match based on how you actually present yourself—not just what you wrote in a profile, but the full picture of who you are.
Dignity-first and privacy-centric. You can tell an AI things you would never say publicly. Preferences you haven't fully articulated. Concerns you wouldn't want attached to your name. Feedback you'd be embarrassed to share with a shadchan. Our system takes all of this into account—and none of it is exposed to anyone else. The AI works for you, learns from you, and keeps what you share private.
Precision from small data. The core innovation addresses the central challenge head-on: generating highly relevant matches from limited information. Where conventional systems need tens of thousands of interactions to detect patterns, ours achieves precision within a handful of attempts. This isn't incremental improvement—it's a different category of capability entirely.
Beyond the Match
Here is perhaps the most important point, and one we want to emphasize clearly: AI matching, however sophisticated, plays only a small role in the shidduch process.
A match suggestion is merely a beginning. What happens next—the introduction, the communication, the navigation of early conversations, the guidance through uncertainty and vulnerability—this is where relationships are actually built or lost. Technology that focuses solely on the initial match misses the vast majority of where help is truly needed.
This realization led us to develop something far more ambitious: an end-to-end AI system that supports the entire shidduch process from initiation to engagement.
How it works.
It starts with discovery. The AI searches across profiles—not just matching keywords, but understanding meaning, factoring in attraction, weighing the dimensions that actually matter. It surfaces suggestions that traditional systems would never find.
Then comes the explanation. Every suggestion includes a rationale—not just who, but why. What do these two people share? Where do their values align? What patterns suggest compatibility? No more "trust me" or "I have a feeling." Real reasons.
The AI initiates outreach on your behalf. It sends the suggestion, follows up, asks the right questions, and collects responses. When the answer is no, it processes the feedback and moves forward. When the answer is yes, it notifies the shadchan to step in.
It coordinates logistics. Our Date Planner draws from over a thousand curated locations, suggesting venues based on preferences and past experiences. It handles the back-and-forth so people can focus on the date itself.
After each date, the Dating Log captures feedback anonymously. Over time, the AI helps you reflect on your dating history—surfacing patterns you may not have noticed, tendencies you've developed, dynamics that keep recurring. It fosters self-awareness, even when that awareness is uncomfortable to confront.
And the Mirror reveals what others see: who you attract, who is attracted to you, why people are saying yes, why they're saying no. This information—so rarely available and so critically important—becomes visible.
Throughout all of this, every AI feature works together. They share context, learn from each interaction, and coordinate seamlessly. This isn't a collection of disconnected tools. It's a unified system designed to support the entire journey.
We currently have over 25 AI features live on the platform, with more coming over the next few months.
Feedback Without Lashon Hara
There's a problem that has haunted shidduchim for as long as anyone can remember: feedback.
When someone says no—whether at the profile stage or after a date—all you get is a binary signal. Yes or no. But a yes or no tells you almost nothing. Why did they decline? What was missing? What would have made it a yes?
This information could dramatically improve future suggestions. But asking for it creates a real problem.
Articulating exactly why you said no invites lashon hara. After a date especially—what benefit does a shadchan derive from hearing negative details about another person? The shidduch is already over. And collecting critical feedback about individuals, feedback that might color how they're perceived in future suggestions, crosses into territory we have no interest in entering.
Our goal is to help everyone find their match. Not to accumulate files of people's shortcomings.
But here's what changes with AI.
You can tell an AI things you would never say publicly. The preferences you haven't fully articulated. The concerns you wouldn't want attached to your name. The feedback you'd be too embarrassed to share with a shadchan—or even admit to yourself.
The AI doesn't store this as gossip. It extracts the underlying patterns and refines its understanding of what you're actually looking for. Your feedback shapes your own recommendations—and never becomes a record of someone else's flaws.
For Shadchanim
AI plays an equally central role on the shadchan's side, beginning with a problem that has plagued matchmakers for as long as shidduchim have existed: communication chaos.
The reality of a working shadchan's life is calls, text messages, WhatsApp threads, emails—scattered across platforms, buried in inboxes, impossible to track. Information gets lost. People get forgotten. Resumes fly back and forth, often outdated before they arrive. It is a hectic, fragmented system held together by memory and good intentions. An AI assistant that consolidates all of this—organizing communications, surfacing what matters, ensuring no one falls through the cracks—provides immediate, tangible relief.
But there is a deeper problem, one we have heard over and over across fourteen years: shadchanim are burned out.
The arithmetic is brutal. Eighty to ninety percent of suggestions are rejected outright. A shadchan invests time understanding two people, sees potential, makes a thoughtful suggestion—and receives a no, often with little explanation. Multiply this across dozens of singles, week after week, year after year. The emotional toll is immense. Many talented shadchanim simply stop making suggestions. They remain nominally active but have lost the energy to keep proposing ideas that will almost certainly be declined.
Our system changes this entirely.
Before a shadchan sends a suggestion, they can run an AI analysis to anticipate challenges, surface potential objections, and identify what to emphasize—increasing the likelihood of a yes before the suggestion is even made.
The work that everyone hates—the back-and-forth, sending resumes, coordinating logistics, collecting feedback, dealing with the nos—AI handles all of it. The shadchan steps in when the answer is yes, when guidance is needed, when the human element becomes essential. This is where a shadchan's true value lies. This is where they should be spending their time.
The End of the Resume
For decades, the shidduch resume has been the central artifact of the process. A sheet of paper listing family background, schools attended, references to call. It has become so ubiquitous that many cannot imagine shidduchim without it.
But consider what a resume actually provides. A generic, boilerplate document—the same format, the same categories, the same carefully curated presentation that reveals little about who someone actually is. Resumes are filled with biases. They encode assumptions about what matters before a single conversation has taken place. They are cumbersome to manage, often outdated, and they reduce human beings to a checklist of externalities.
This isn't a hiring decision. It's a life decision.
The future of shidduchim isn't about better resumes. It's about AI that gets to the heart of what actually makes or breaks a relationship—the nuances, the subtleties, the things that matter but never appear on a form.
Instead of generic templates, information should be specific, relevant, and oriented toward understanding. What does this person share with you? Where do your values align? What patterns suggest compatibility? This is a far more effective way of conveying personal information than a boilerplate sheet of paper that has served the community more out of habit than effectiveness.
People may still call references, and we encourage them to do so. But the resume itself? We see a future where it fades from its central role. Not because information doesn't matter—but because there are far better ways to convey it.
Smart Subscription
We recognize that people are fatigued. Fatigued from sharing their information on multiple systems. Fatigued from the expectations, the disappointments, the anxiety of waiting. We want to alleviate that.
AI matching is fundamentally different from any other platform. Where other platforms advocate for more—more matches, more dates, more activity—we have always focused on less. Quality over quantity. This has been our philosophy since founding, and it's consistent with decades of social psychology research: more is not better. More is overwhelming. More leads to worse decisions.
We never guarantee a specific number of dates. We never guarantee a specific number of matches. What we guarantee is that our AI system will find the best match possible based on your profile.
It's important to emphasize: based on your profile. Profiles may be outdated, worded incorrectly, or perceived differently than intended. Many can benefit from refinement—and our system can help with that too.
Our AI will not suggest people who are actively dating someone else. It constantly adjusts based on who is available and ready.
Here's how it works: you sign up, and you're only billed when we have high-quality AI matches for you that month. If you've exhausted the pool, or if the people most relevant to you are busy dating or traveling, we automatically pause your account. No need to worry about subscribing and unsubscribing. No need to manage anything.
Our goal—our business model, really—is to bring as many people as possible into the system and to remove as many people as possible from our database by leading them to engagement.
This automatic system of activating when you're ready for shidduchim saves you from the cycle of expectation and disappointment, from managing multiple subscriptions, from wondering if you're wasting money on a service that has nothing for you.
For $9 a month, this is by far the best value anyone could expect from a system like this. And the value will only grow with time.
A New Chapter for Shidduchim
What we have described is not a vision for the future. It is what exists today.
The technology is built. The platform is live. Over 25 AI features are working together—matching with unprecedented precision, supporting singles through the process, freeing shadchanim from the burden that has been crushing them for years.
For too long, the shidduch world has been caught between two unsatisfying realities: the traditional approach, with all its inefficiencies and exhaustion, or technological promises that never delivered. We refused to accept that these were the only options.
We spent years understanding why existing solutions failed. We developed novel techniques where none existed. We rebuilt everything from the ground up—hundreds of thousands of lines of code—to create something entirely new.
The result is a system where shadchanim can help more people. Where singles are supported at every step. Where AI handles the burden so people can focus on what only people can do.
The shidduch crisis is real, and it will not be solved by technology alone. But after fourteen years of this work, we believe that technology built the right way, for the right reasons, within the real constraints of shidduchim, can make a profound difference.
This is what we have built at Yismach.
And we're just getting started.