zerosumalpha

I build the systems that execute trades. Infra, risk engines, and what AI actually changes — where mistakes cost real money. The engineering behind the trade.

[zerosumalpha @zerosumalpha]

Place a limit buy on Binance: the field is `quantity`.

On OKX: `sz`.
On Hyperliquid: `s`, the asset is an integer, and the side is a boolean.

Exchange connectivity isn't about calling an API. It's making your system speak all of them.

[zerosumalpha @zerosumalpha]

Exchange connectivity isn't build-once. New listings need symbol refreshes. API updates silently break your parsers.

Full post with comparison tables: https://t.co/5kV4GRn0h4

What's the worst exchange API quirk you've hit?

[zerosumalpha @zerosumalpha]

Place a limit buy on Binance: the field is `quantity`.

On OKX: `sz`.
On Hyperliquid: `s`, the asset is an integer, and the side is a boolean.

Exchange connectivity isn't about calling an API. It's making your system speak all of them.

[zerosumalpha @zerosumalpha]

OKX returns timestamps as strings. Every other exchange returns integers.

In Go, your int64 won't unmarshal from a JSON string. You find out the hard way.

Binance gives you a server timestamp. OKX and Hyperliquid don't. Without one, you can't separate network latency from exchange processing time.

[zerosumalpha @zerosumalpha]

An LLM strategy loses money from a WebSocket drop or a missing kill switch — everyone blames the model. The execution layer failed beneath it. The gap in agentic trading isn't the models. It's the runtime.

[zerosumalpha @zerosumalpha]

@MINHxDYNASTY you never know until you try

[zerosumalpha @zerosumalpha]

@ff_v12 not your keys, not your crypto

[zerosumalpha @zerosumalpha]

@hamptonism supplements could help, look up Magnesium L-threonate , it helps with good sleep and focus

note: not medical advise, please check with you doctor

[zerosumalpha @zerosumalpha]

@jussy_world learning & building agentic infrastructure and prepare for autonomous agentic future

[zerosumalpha @zerosumalpha]

@0xTDG what financial products are you looking to build?

have a vision for it and a rough product spec and just build it . fine tune along the way.

you cannot build something perfectly on day 1.

ai tools definitely helps

[zerosumalpha @zerosumalpha]

@CodeWithAmann solo building

[zerosumalpha @zerosumalpha]

Trade surveillance has a false positive problem.

The fix being tested at major banks: agents that correlate relationships across orders, timing, and market context — not single-event rule triggers.

Human reviewers still make the call. The open question is the audit trail when the agent misses something.

[zerosumalpha @zerosumalpha]

@iamdisplacement Buy low sell high

[zerosumalpha @zerosumalpha]

Consolidated risk across dozens of exchange accounts sounds like an aggregation problem. It's not. The engineering works, the tests pass. But whether the numbers are actually correct? That needs a trader staring at it saying "that exposure doesn't look right."You can't unit test domain knowledge.

[zerosumalpha @zerosumalpha]

Most crypto volume is already automated — bots, algos, market makers. But the next wave is AI agents making real decisions. And they're going to hit the same wall every algo trader hits: the gap isn't intelligence, it's infrastructure. Execution quality, risk management, state tracking, reconciliation. The stuff between "the agent decided to buy" and "the trade actually happened.

[zerosumalpha @zerosumalpha]

DeFi protocols are building agent-native interfaces now. @1inch just shipped MCP integration, 15 DeFi APIs, all swap types, cross-chain, gasless. Single-line agent setup. The protocols are reshaping around the agents, not the other way around.

[zerosumalpha @zerosumalpha]

Testnet doesn't make a trading system bulletproof. Prod does. Real rate limits, real liquidity gaps — exchange quirks that only show up on real accounts with real money. You learn what breaks by breaking it.