Leo 👨🏻‍💻

CTO @DeSpreadTeam | Prev. SWE @LINEjp_official

[Leo 👨🏻‍💻 @unexpected_leo]

I'm excited to share that @DeSpreadTeam took 3rd place for the @AptosLabs Incentivized Testnet 3 and got a chance to participate as validator in upcoming Aptos Mainnet!

Thread 👇
https://t.co/CiSmOoAogk

[Leo 👨🏻‍💻 @unexpected_leo]

🔑 Scale & source

Add schemas + auto metrics and this becomes CI-your LLMs improve while you sleep.

Prompt engineering is the gearbox for enterprise‑grade GenAI.

Full paper 👉 https://t.co/9i0awtZhts

#PromptEngineering #LLM #GenerativeAI #LangChain #AI

[Leo 👨🏻‍💻 @unexpected_leo]

🧑‍🍳 Quick recipes

- Shuffle label order in Few‑Shot to kill bias
- Temp 0 for CoT and parse only the “Answer:” line
- LangChain lines spin up a ReAct agent
- Schedule a weekly APE loop to keep prompts sharp

[Leo 👨🏻‍💻 @unexpected_leo]

🚀 Real‑world impact

- Old way : hand‑tune prompts for every task.
- Playbook way : one template reused across models, evaluation fully automated. Less code, more outcome.

[Leo 👨🏻‍💻 @unexpected_leo]

5. Automatic Prompt Engineering (APE) : the model drafts 10 candidate prompts, scores them with BLEU/ROUGE/F1, and ships the winner-prompt tuning in minutes, not days.

[Leo 👨🏻‍💻 @unexpected_leo]

3. CoT + Self‑Consistency : add “Let’s think step by step,” sample 5 runs, keep the majority answer-logic slips vanish.
4. ReAct agents : Thought → Action → Observation loops (LangChain) let the model call live tools like Search, SQL, or Python.

[Leo 👨🏻‍💻 @unexpected_leo]

⚒️ Toolkit

1. Zero/One/Few‑Shot : feed 0‑3 balanced examples; our email classifier jumped +15 % F1.
2. Step‑Back : ask for high‑level principles first, then the answer-ideas get richer and less cliché.

[Leo 👨🏻‍💻 @unexpected_leo]

💡 Why it matters

Prompting used to be pure guesswork.

The playbook turns it into repeatable patterns that boost accuracy and creativity in hours, not weeks.

Finally, a checklist-not a dartboard.

[Leo 👨🏻‍💻 @unexpected_leo]

🚀 Just blasted through @Google new Prompt Engineering white paper.

5 quick wins you can swipe for your next sprint.

Here’s what you need to do👇

[Leo 👨🏻‍💻 @unexpected_leo]

- Smithery hosted version: https://t.co/DNEFERi48P
- GitHub: https://t.co/FkzUMV2iaH

Happy building 🔥

#AI #MCP #Metabase

[Leo 👨🏻‍💻 @unexpected_leo]

🚀 Excited to share my MCP project.

My open-source implementation of
@metabase
MCP server from scratch.

I frequently use Metabase but couldn’t find a suitable MCP, so I built one myself.

[Leo 👨🏻‍💻 @unexpected_leo]

In the project, I implemented the following features:

Bridge AI & Metabase with 'metabase://' URIs, session/API auth, and natural language. Run dashboards, saved questions, or custom SQL. Get structured JSON output.

[Leo 👨🏻‍💻 @unexpected_leo]

@MoneyStack9 Money!

[Kevin Kern @kevinkern]

A new Cursor Update is rolling out! (Version 0.47)

Click to see the whats new 🧵👇

[Leo 👨🏻‍💻 @unexpected_leo]

따로 보안이 필요없는 외부 데이터를 인풋으로 하는 MCP는 크게 효용성이 있는지는 아직 의문

[Leo 👨🏻‍💻 @unexpected_leo]

MCP 써본 후기

기존에도 RAG, Function Calling, Agentic AI 등으로 구현 가능했지만, MCP는 이를 표준화 & 구조화 함으로써 구현 레벨에서 더 쉽게 AI Agent를 만들 수 있음

로컬에서 샘플링 데이터로 테스트 해본 결과 유의미한 결과가 나오는 것을 확인함

[Alex Xu @alexxubyte]

What is MCP?

Why is everyone talking about it?

Let’s take a closer look.

Model Context Protocol (MCP) is a new system introduced by Anthropic to make AI models more powerful.

[Leo 👨🏻‍💻 @unexpected_leo]

주의할점

MCP 보안 레이어가 완벽하지 않아서 도입 전 테스트가 필요함
Private 오픈소스 LLM을 사용하면 괜찮을 것 같다

결국 MCP를 엔터프라이즈로써 잘 활용하려면 오픈소스 LLM을 잘 구축놔야할 것 같음

[Leo 👨🏻‍💻 @unexpected_leo]

MCP 서버 리스트
https://t.co/Up7THf4d8b
https://t.co/lK3Pzon3qD
https://t.co/AEA7qKO8qf

[Leo 👨🏻‍💻 @unexpected_leo]

개인이나 회사의 데이터를 가지고 어떻게 효율적으로 사용할지 고민 중이었는데 MCP로 재밌는 것들을 해볼 수 있을 것 같음

AI를 사용하기 점점 더 쉬워지는 시대가 오는것 같다

[John Rush @johnrush]

What is MCP & why it's a big (huge) deal:

(model context protocol)

TLDR: MCP makes it possible for AI Tools to use external tools. E.g. Chatbot/IDE/AI-Agent can use Gmail/GoogleDrive/WeatherApp etc.

Detailed explanation for both, tech & non tech people (+demos):

1) AI Tools (chatbots, wrappers, agents, code generator, etc) wanna talk to external systems.

In pre-MCP world, one would have to write code to connect AI tool to the external system via API. Which meant every connection had to be pre-coded.

It also meant that every AI tool had to hard code its connection to every other tool. So if there are 1000 AI tools and 1000 external tools, then 1000000 hard-coded connections via API.

2) MCP is a standard protocol. This means that every AI tool has to implement this once, and then it can connect to thousands of external tools via this protocol.

3) The same goes for external tools. They all have to create an MCP server just once, and all AI tools that support MCP can connect to them.

4) It's a huge deal. Imagine 10k AI tools and 10k external tools now all have to implement MCP just once each. So it's 20k implementations. Versus 10k*10k=100M implementations.

5) This whole thing can also run on the cloud or on local computer.

See demos:

[Jeffers🀄️ @Jefferstellar]

MONAD TESTNET ANALYSIS 📊

For a long time I have been wanting to combine my data science skills and my love for @monad_xyz . Analyzing testnet data using @flipsidecrypto was a perfect way for me to do this.

For this thread I analyzed native MON transfers between two EOAs (no smart contract interactions). We only consider data since the testnet went live so the analysis reflects organic usage of the testnet.

No bias just data, here’s what I found 🧵👇