A table tennis AI???

This user has no status.
This user has no status.
Member
May 2022
15
4
35
Hello guys and gals!
I've developed an AI Agent for table tennis, based on the experience of the best coaches I've met and read (two brazillian coaches, one american and one chinese). It's working very well and we're using in our local club. It's wonderful!
My question is: Do you guys think it would be a good tool? Do you guys would pay for it? (Not much, just because I have to pay for running the application) Do you think it would be a good thing and ppl gonna use it?


It's focused on tactics and strategic development and differently from some coaches, it can give personalized tips: f.e. "I'm a backhand counter-driver dominant close to the table, how can I be better? What shots do I have to develop to be stronger?" I had a coach who told me: don't hit backhand in the middle of the table, attack only with forehand, don't focus on BH, only develop a powerful forehand loop, etc... He is a good coach, but he has a strong opinion about dominating with backhand and couldn't give me personalized tips for my development. But this AI could give me precise tips: "
  1. Service Tactics:
  • Use a variety of spin services like backspin and sidespin to disrupt your opponent's rhythm and attack.
  • Serve mostly inverted-pendulum and hook serves (right-side spin) to use your backhand earlier in the game.
  • Serve short to the forehand side to force weaker returns that you can attack with your strong backhand.
  1. Rally Tactics:
  • Look to counter-drive and attack with your backhand whenever possible, especially against underspin balls.
  • Develop a solid forehand loop to complement your backhand and attack from both wings.
  1. Mentality:
  • Stay patient and look for the right opportunities to unleash your backhand counter-drives.
  • Don't get impatient and go for too many risky winners from difficult positions.
  • Have confidence in your backhand strengths close to the table.
  1. General Tips:
  • Work on your footwork to quickly get into an ideal backhand stance close to the table.
  • Improve your consistency by doing multiball backhand counter-driving drills.
  • Develop a strong backhand banana flick to change spin and throw off timing.
  • Consider learning a potent forehand loop kill against backspin to add variation
"
 
This user has no status.
This user has no status.
Member
May 2022
15
4
35
if this app can watch me play and call timeouts or give me court wide coaching during the breaks, just like a top international coach, then you got something.
Maybe in a v2 of the app hahaha
It's possible using OCR, but the main problem here is: How strong is your smartphones speakers?
 
This user has no status.
Hello guys and gals!
I've developed an AI Agent for table tennis, based on the experience of the best coaches I've met and read (two brazillian coaches, one american and one chinese). It's working very well and we're using in our local club. It's wonderful!
My question is: Do you guys think it would be a good tool? Do you guys would pay for it? (Not much, just because I have to pay for running the application) Do you think it would be a good thing and ppl gonna use it?


It's focused on tactics and strategic development and differently from some coaches, it can give personalized tips: f.e. "I'm a backhand counter-driver dominant close to the table, how can I be better? What shots do I have to develop to be stronger?" I had a coach who told me: don't hit backhand in the middle of the table, attack only with forehand, don't focus on BH, only develop a powerful forehand loop, etc... He is a good coach, but he has a strong opinion about dominating with backhand and couldn't give me personalized tips for my development. But this AI could give me precise tips: "
  1. Service Tactics:
  • Use a variety of spin services like backspin and sidespin to disrupt your opponent's rhythm and attack.
  • Serve mostly inverted-pendulum and hook serves (right-side spin) to use your backhand earlier in the game.
  • Serve short to the forehand side to force weaker returns that you can attack with your strong backhand.
  1. Rally Tactics:
  • Look to counter-drive and attack with your backhand whenever possible, especially against underspin balls.
  • Develop a solid forehand loop to complement your backhand and attack from both wings.
  1. Mentality:
  • Stay patient and look for the right opportunities to unleash your backhand counter-drives.
  • Don't get impatient and go for too many risky winners from difficult positions.
  • Have confidence in your backhand strengths close to the table.
  1. General Tips:
  • Work on your footwork to quickly get into an ideal backhand stance close to the table.
  • Improve your consistency by doing multiball backhand counter-driving drills.
  • Develop a strong backhand banana flick to change spin and throw off timing.
  • Consider learning a potent forehand loop kill against backspin to add variation
"
How is this product different from other, free alternatives such as DeepSeek, ChatGPT, Perplexity and etc.? Because when you formulate the questions like in this example, they give practically identical answers.
My question: what is so special about this product that you should pay for it?
 
This user has no status.
This user has no status.
Member
May 2022
15
4
35
How is this product different from other, free alternatives such as DeepSeek, ChatGPT, Perplexity and etc.? Because when you formulate the questions like in this example, they give practically identical answers.
My question: what is so special about this product that you should pay for it?
That's an excellent question, Taute!
When you want to give precise answers, you have to do a fine-tuning of a pre-existing model OR a technique called RAG (Retrieval-Augmented-Generation), which are ways to improving the models response accuracy.
The answers of vanilla models are not precise most of the time, they don't have a curated reference for table tennis, and on the technical side they have higher temperature (a term to define the randomness of an AI text).
When we develop an specific AI Agent we have to make sure the temperature is low or zero, and we have a curated and very strong reference for the AI. Doing so, we have precise responses, instead of generic ones.
This one I developed for our club is a fine-tuning AND a RAG, meaning it will give precise answers 100% of the time based on the provided context (which is the way I can improve my app, gathering better and more coaching data).
 
This user has no status.
That's an excellent question, Taute!
When you want to give precise answers, you have to do a fine-tuning of a pre-existing model OR a technique called RAG (Retrieval-Augmented-Generation), which are ways to improving the models response accuracy.
The answers of vanilla models are not precise most of the time, they don't have a curated reference for table tennis, and on the technical side they have higher temperature (a term to define the randomness of an AI text).
When we develop an specific AI Agent we have to make sure the temperature is low or zero, and we have a curated and very strong reference for the AI. Doing so, we have precise responses, instead of generic ones.
This one I developed for our club is a fine-tuning AND a RAG, meaning it will give precise answers 100% of the time based on the provided context (which is the way I can improve my app, gathering better and more coaching data).
Interesting. I think I would definitely try this product :)
 
  • Like
Reactions: Gkustodor
This user has no status.
This user has no status.
Member
May 2022
15
4
35
For example, I asked "how to play against hitters?"

Answer:
"Let me help you understand how to approach playing against hitters.

Based on the context, hitters have some specific characteristics and weaknesses we can exploit:

  1. Middle Play: Hitters are often weak in the middle (elbow area) because they play close to the table and have little time to react. Use this to your advantage by attacking their elbow area frequently.
  2. Serve Strategy:
  • Use deep backspin serves, as hitters often struggle with looping these effectively
  • Mix in long serves that force them to make decisions
  • Vary your serves to prevent them from getting into a hitting rhythm
  1. Rally Tactics:
  • Keep them off-balance by mixing up the pace
  • Don't give them easy balls to smash
  • Force them to loop rather than hit when possible
  • Use good placement to make them move, as they prefer to stay close to the table
  1. Counter their Aggression:
  • Remember that hitters are looking to either end the point or force a weak return they can smash
  • Stay patient and don't let their aggressive style rush you into mistakes
  • Use good quality loops to keep them at bay
The key is to disrupt their preferred close-to-table hitting game and force them into uncomfortable positions where they can't execute their powerful shots effectively."
 
This user has no status.
This user has no status.
Active Member
Apr 2013
710
603
1,586
Read 1 reviews
For example, I asked "how to play against hitters?"

Answer:
"Let me help you understand how to approach playing against hitters.

Based on the context, hitters have some specific characteristics and weaknesses we can exploit:

  1. Middle Play: Hitters are often weak in the middle (elbow area) because they play close to the table and have little time to react. Use this to your advantage by attacking their elbow area frequently.
  2. Serve Strategy:
  • Use deep backspin serves, as hitters often struggle with looping these effectively
  • Mix in long serves that force them to make decisions
  • Vary your serves to prevent them from getting into a hitting rhythm
  1. Rally Tactics:
  • Keep them off-balance by mixing up the pace
  • Don't give them easy balls to smash
  • Force them to loop rather than hit when possible
  • Use good placement to make them move, as they prefer to stay close to the table
  1. Counter their Aggression:
  • Remember that hitters are looking to either end the point or force a weak return they can smash
  • Stay patient and don't let their aggressive style rush you into mistakes
  • Use good quality loops to keep them at bay
The key is to disrupt their preferred close-to-table hitting game and force them into uncomfortable positions where they can't execute their powerful shots effectively."
Im sorry but thats not going to work for anyone... Especially assuming the majority of the users wil be beginners and amateurs looking to improve their game. They wont remember all these 'helpful tips'.

What I would be looking for in a tactical AI coach would be to record and analyse my matches and have live feedback to read in between games.

Something like: "You have won the game 11-8, good job!".

Things you do well:
1. "You won 5/5 points directly by slow-looping a push with your forehand."
2. "You won 3/4 points that started by pushing the serve long to the opponents elbow."

Things you could improve:
1. "You lost 4/5 points that started by pushing long to the opponents forehand. Try targeting his elbow more since that won you 3/4 points."
2. "You made 0/3 points when trying to loop aggressively against a long push to the backhand." "Try to loop more controlled with more spin: you won 5/5 points that started with a slow-loop in last game."
"Or if your backhand isnt as secure, try to push more to the opponents middle: you won 3/4 points that started with a long push to the opponents elbow in the last game."

Most effective serve:
"3/4 points where won when this serve was used."


And based on the analysis of multiple matches, it should create a training program with exercises for you to maintain your strengths (the way you generally win most points), and improve your weaknesses (the way you generally lose most points).
 
This user has no status.
This user has no status.
Member
May 2022
15
4
35
Im sorry but thats not going to work for anyone... Especially assuming the majority of the users wil be beginners and amateurs looking to improve their game. They wont remember all these 'helpful tips'.

What I would be looking for in a tactical AI coach would be to record and analyse my matches and have live feedback to read in between games.

Something like: "You have won the game 11-8, good job!".

Things you do well:
1. "You won 5/5 points directly by slow-looping a push with your forehand."
2. "You won 3/4 points that started by pushing the serve long to the opponents elbow."

Things you could improve:
1. "You lost 4/5 points that started by pushing long to the opponents forehand. Try targeting his elbow more since that won you 3/4 points."
2. "You made 0/3 points when trying to loop aggressively against a long push to the backhand." "Try to loop more controlled with more spin: you won 5/5 points that started with a slow-loop in last game."
"Or if your backhand isnt as secure, try to push more to the opponents middle: you won 3/4 points that started with a long push to the opponents elbow in the last game."

Most effective serve:
"3/4 points where won when this serve was used."


And based on the analysis of multiple matches, it should create a training program with exercises for you to maintain your strengths (the way you generally win most points), and improve your weaknesses (the way you generally lose most points).
Thanks, Amayzde! That’s amazing feedback.

I would love an app like that too!
Is it possible with the tech we have now? Yes!
Is it easy? No.
Could I do it? Yes! (But it would take some time to train a computer vision model to identify mistakes and successes in table tennis—it would require hundreds of footage clips from amateur players.)

But I will definitely look into it. It’s the perfect solution.
The real question is: if I get an app like that up and running, would it pay off?
 
  • Like
Reactions: Amayzde
This user has no status.
This user has no status.
Active Member
Apr 2013
710
603
1,586
Read 1 reviews
Thanks, Amayzde! That’s amazing feedback.

I would love an app like that too!
Is it possible with the tech we have now? Yes!
Is it easy? No.
Could I do it? Yes! (But it would take some time to train a computer vision model to identify mistakes and successes in table tennis—it would require hundreds of footage clips from amateur players.)

But I will definitely look into it. It’s the perfect solution.
The real question is: if I get an app like that up and running, would it pay off?
I mean, if the app works as well as expected, and isnt too expensive, it will spread quickly. Im not a marketing expert though.

But you shouldnt develop the app for the sake of it paying off. Thats probably a quick way of setting yourself up for disappointment.

You should develop it because you are enthusiastic about helping people with the technology we have today. Passion will be more effective then pushing for succes.

I am definitely interested in something like this, so I am eager to get it if it comes out.
 
  • Like
Reactions: Gkustodor
This user has no status.
This user has no status.
Member
May 2022
15
4
35
I mean, if the app works as well as expected, and isnt too expensive, it will spread quickly. Im not a marketing expert though.

But you shouldnt develop the app for the sake of it paying off. Thats probably a quick way of setting yourself up for disappointment.

You should develop it because you are enthusiastic about helping people with the technology we have today. Passion will be more effective then pushing for succes.

I am definitely interested in something like this, so I am eager to get it if it comes out.
Cool! I will study the best way of implementing and running an app like this.
The thing about paying off it's not a greedy thing or some kind of money-first mindset haha It's just because running AI Models has a cost. Even worse: dollar cost haha (A dollar = around 5 Reais; Minimum wage is: US$7.25 vs. R$ 8.82 (US$1.8), that's important, cause Web Services, AI Models, charge thinking in the US and EU Markets... so it's very expensive for us)

So, maybe I would need to find some investors and/or make sure there's enough table tennis nerds that would use and pay for this app...

If I was a millionaire, I would definitely do it for free tho hahaha
 
This user has no status.
This user has no status.
Member
Dec 2016
31
31
66
I don't think there's enough utility here until you start to approach some of the capability of what Tony mentioned. I don't think you need to go that far, but async/offline analysis of match/training footage with personalized coaching and technique tips would be the minimum for a paid product.

As it is, the main value prop you have is the dataset of coaching tips that you're fine-tuning/have RAG implemented on top of. Unless you've got an army of very high level coaches generating this dataset, it's relatively easy for someone to replicate this by scraping ie, TTD or transcripts of the thousands of YouTube videos. Of course that takes time, but a lot of that stuff is in the public domain and has video to supplement.

Also, it assumes that someone is able to accurately formulate or even understand the problem they have. If you're wondering how to play against a specific type of player (at the granularity of "chopper", "hitter", "looper"), I'm not sure that target demographic would pay much more than a dollar or two, if that. And that's also a pretty small addressable market.

If you're serious about pushing the technical capabilities of something like this, I'd recommend open-sourcing the dataset so other people can contribute to it, or self-host if they want/can. For the rest, they might be willing to throw a couple dollars your way for the convenience.
 
This user has no status.
This user has no status.
Member
Aug 2022
223
340
566
Don't do this with the expectation of making $$$, because you probably will not in the table tennis market. Do it because it's an interesting way to develop AI and may get you a good job.

Truls has been sponsored by Stupa Analytics which has an AI table tennis app. Stupa appears to have many features that people have mentioned. Have you compared what you are planning to Stupa? It should be easy for Stupa to add what you are doing to their existing product.
 
This user has no status.
This user has no status.
Member
May 2022
15
4
35
Thank you all for the feedback! Honestly, I started developing this without any expectations of making money... the whole idea came from my own experience playing at small clubs without proper coaching. In countries like Brazil, for example, top-level coaches are concentrated in big cities like São Paulo, while players from smaller towns often hit a plateau because they lack tactical and strategic guidance. This MVP was built with them in mind, to give players access to insights that could help them improve, even without a high-level coach around.

So far, the system has been effective in that niche, but I understand that scaling it further requires market research and more advanced features.

Regarding learning from conversations: yes, it's totally possible for the AI to retain context and improve over time in a user-based scenario. I've been exploring that as a next step, since making it learn from all conversations would probably pollute the dataset. Letting the model adapt to each user's interactions individually seems like a better practice for this kind of application.

Again, I really appreciate all the input – I'm definitely going to find ways to push the project forward.
 
  • Like
Reactions: haggisv
Top