Himanshu sharma

FlutterDevrel | #BTCSL W24

[Himanshu sharma @Himansh68]

@shl One agent to rule them all — until context overflow hits.

[Himanshu sharma @Himansh68]

Finally, safety checks and alignment systems help ensure the response follows instructions and avoids harmful or misleading outputs.

The biggest misconception?

LLMs don’t search a giant database of answers.

They predict the most likely next token based on:
• Training
• Context
• Retrieved knowledge (RAG)
• Tool outputs

That’s what makes modern AI assistants so powerful.

#AI #LLM #Claude #Transformers #RAG #VectorDatabase #MachineLearning

[Himanshu sharma @Himansh68]

Most people think Claude (or any LLM) “knows” the answer.

It doesn’t.

Here’s what actually happens behind the scenes 🧵👇

[Himanshu sharma @Himansh68]

If connected to tools, the model may use RAG (Retrieval-Augmented Generation):

Question → Embedding → Vector Database → Retrieve relevant documents → Add context → Generate grounded answer.

[Himanshu sharma @Himansh68]

const constructors in Flutter aren't optional. They're the difference between 40fps and 60fps. Use them everywhere you can.

#flutter

[Himanshu sharma @Himansh68]

@ayanamindes Let’s connect

[Himanshu sharma @Himansh68]

@VivienMahe Flutter

[Himanshu sharma @Himansh68]

@shl You can vibe code the product.

You can’t vibe code product-market fit.

And you definitely can’t vibe code paying customers.

[Himanshu sharma @Himansh68]

@melaniecarstens @gzani @VoltFiLabs 🔥

[Himanshu sharma @Himansh68]

@trikcode Hi

[Himanshu sharma @Himansh68]

@mischavdburg Devops

[Himanshu sharma @Himansh68]

@trikcode Hi

[Himanshu sharma @Himansh68]

@tonklaus Let’s connect

[Himanshu sharma @Himansh68]

@shl Still a job that AI can’t see

[Himanshu sharma @Himansh68]

@shl You don’t need to be a math genius to code well.
You need clarity of thought, structured logic, and practice.

[Himanshu sharma @Himansh68]

@shl It’s a product fit market