The problem Nexora solves

DeFi lets everyone use the same leverage — a brand new trader and someone with years of experience both get 20x if they want it. Nobody's checking if that's actually a good idea for them. Turns out it isn't; new traders get liquidated way more often because nothing ever holds them back.

The obvious fix — let the platform check your trading history — means putting your entire trade record on a public ledger forever. Most people won't accept that trade-off.

What Nexora does
Looks at how you actually trade and sorts you into Novice, Balanced, or Expert
Turns that into real limits — a smart contract enforces them, not just a UI suggestion
Your raw trading history stays private; only the result goes on-chain
Who it's for
New traders who'd rather be held back a little than blow up their first position
Experienced traders who want their track record to actually count for something
Protocols that want risk-based tiers without forcing users to dox their history

Basically: personalized leverage limits, without giving up your privacy to get them.

Challenges we ran into

Picking the right model. Started with a Decision Tree — worked, but accuracy sat around 75%. Switched to Random Forest, first with a capped depth, which got to ~85%. Removed the depth limit and landed at ~92.6%. Wasn't obvious upfront which would work best, so we just tried each one and kept what performed.

Figuring out what "private" actually means here. Building real zero-knowledge cryptography from scratch wasn't realistic in the time we had. So instead of pretending we'd built a full ZK-SNARK circuit, we built a hash-based attestation — the risk result gets hashed, and that hash is what goes on-chain, not the raw trading data. It's a stand-in for where a real proof system would sit, not the real thing, and we're upfront about that.

Foundry silently building nothing. forge build kept saying "Nothing to compile" even from the right folder. No foundry.toml, so Forge was looking for source files in completely the wrong place. One config file fixed it.

forge install failing without telling us. It just didn't install anything, no error shown. Worked around it by pulling the library in directly with git clone instead.

Buttons that looked broken but weren't. Backend logs showed every click working. Turned out a CSS rule was quietly keeping the home screen visible no matter what, so the dashboard was rendering underneath it, invisible. Fixed with one CSS rule.

Two traders sharing one identity. Both demo personas were signing from the same blockchain account early on, so their balances and positions kept colliding. Gave each one its own account.

Actually connecting everything. The model, the attestation, the contract, Anvil, and the frontend were all built as separate working pieces — getting them talking to each other as one flow was its own project. Model output had to match what the policy layer expected, the policy had to match what the contract accepted, and the frontend had to poll all of it live without breaking. Most of the bugs here weren't in any single piece — they showed up only once two pieces were forced to agree with each other.