We Just Made a Big Upgrade to Chessalyz.ai's Feedback Algorithm
How we made the AI feedback feel more like a real coaching conversation
It’s hard to believe that Varun and I started working on Chessalyz.ai almost a year ago now. Back in February 2025, we had an idea for a tool that could help players analyze their games in a more meaningful way. Not just “here’s what the engine says” but actually walking through your thinking and getting feedback on your decision-making process.
A lot has changed since then.
I’ve mentioned Chessalyz in a few articles here and there, but I wanted to dedicate a full post to something we shipped this week that I’m really excited about: a major upgrade to the AI feedback algorithm.
For those unfamiliar with how Chessalyz works, here’s the basic loop: You upload a game, and the AI coach (we named it Matty) asks you questions about your critical moments. You type out your reasoning, and then you get feedback from Matty based on your answer. That back-and-forth is the core of what makes the tool useful for improvement. It’s not about memorizing what the engine thinks is best. It’s about understanding your own thought process and learning how to think better at the board.
The previous feedback was solid and helped a lot of users work through their games, but we knew there was room to make it even better.
Sometimes Matty would tell you that your move “didn’t significantly improve your position” without fully unpacking why. It would suggest a better move but the explanation could feel a bit surface-level. And it had a tendency to lead with phrases like “It’s great that you’re thinking about...” which, while well-intentioned, didn’t always feel like how a coach would actually talk to you over the board.
The new feedback is different. Let me walk you through what changed:
It actually engages with your thinking first
Matty our AI coach now does a better job of acknowledging what you were trying to do before offering corrections. If you played a prophylactic move to stop a knight jump, it recognizes that logic. If you grabbed a pawn because you wanted material, it understands that instinct. Then it explains why something else might have been stronger.
This is how real coaching conversations work. A good coach doesn’t just tell you that you’re wrong. They first make sure they understand what you were going for, and then they redirect you from there.
It explains the “why” behind the correction
The new feedback goes deeper into the reasoning. Instead of just pointing you toward a better move, it tells you things like “you spent a tempo on a threat that isn’t immediately dangerous” or “you let Black breathe for a moment instead of keeping the initiative.”
These explanations actually teach you something about the position. You walk away understanding not just what was better, but why it was better and what you missed in your original thinking.
It gives you thinking frameworks you can use in future games
This is the part I’m most excited about.
The new feedback includes questions you can ask yourself at the board. Things like “Do I have a forcing move here (check, capture, threat)?” or “Is there a more concrete way to use my pieces right now?”
These aren’t just answers to one specific position. They’re tools you can bring to your own games. The goal has always been for Chessalyz to help you become a better thinker at the board, not just to tell you what Stockfish prefers. This update moves us a lot closer to that goal.
It sounds like a coach sitting across from you
The tone feels more direct and conversational. More like actual analysis you’d get from a training partner or instructor. Less robotic, more human.
Try it out and let me know what you think
Here’s where I have a small ask for you.
If you’ve been curious about Chessalyz or haven’t checked it out in a while, I’d really appreciate it if you gave the new feedback a try. Head over to Chessalyz.ai, create an account, then connect your Chess.com/lichess.org account (or copy in a PGN), select a game, click “Chessalyz with me!”, answer Matty’s questions like you normally would, and pay attention to how the feedback feels.
Then shoot me a message or leave a comment and let me know what you think. Does it feel more helpful? More like a real coaching conversation? Is there something we’re still missing?
Varun and I have been heads-down on this project for almost a year now, and feedback from real users is genuinely the most valuable thing we can get. It helps us know what’s working and what to focus on next.
Join me for a live Chessalyz session tomorrow:
If you want to see the new feedback in action, I'm hosting a live game analysis session tomorrow (Saturday, January 17th) at 11am Eastern on Zoom. I'll be using Chessalyz.ai to break down real games from the community, move by move. It's a great chance to see how the tool works and how the new feedback feels in practice. You can join using this link:
https://us02web.zoom.us/j/81407250165?pwd=sSqtofqv45tjUW4m0d2CsCukC8tAoo.1
Meeting ID: 814 0725 0165
Passcode: chess
If you want your game featured, drop a link in the comments and I'll pick a few of the most interesting ones to analyze live. Hope to see you there and happy analyzing!
Generally I like what you have made and find it nice to watch and use.
Hovewer, AI explanations are misleading sometimes. I had a wild tactic game which has several unclear paths to follow. e.g. my opponent takes on f3 and I have options to take with pawn gf3: or with queen. Anyway I am ruining pawn structure on king side and my king is stil in center. I worry about its security so I offer exchange of queens with Qf3: and opponent accepts and my big advantage goes to zero. It is not best move because there exists a tactic with open check to his king and simultaneous attack on his queen.
I would like to see some explanation in that direction. Instead I got this: "It's great to see you thinking about king safety and the impact of major pieces like queens. However, by exchanging queens with Qxf3, you missed an opportunity to reinforce your pawn structure with gxf3. This move would have kept more material on the board and allowed you to later control the center with f3 and e4 (there is no pawn on e line, just doubled isolated f2/f3). Remember to consider how pawn moves can enhance your position and maintain dynamic flexibility. Try exploring the position with gxf3 and observe how it shapes the game."
Can you comment on this? I guess it is hard to make AI analyze set of moves and grasp real idea behind moves. How do you tackle this problem? I think this is 'make it or break it' point of your AI implementation.
Check out this #chess game: Fatima vs Antennaman - https://www.chess.com/game/computer/527905765