Drag-and-drop UI to build AI agents!
Langflow is a powerful visual tool for building and deploying AI-powered agents and workflows - without writing any code.
Supports all major LLMs, vector DBs, etc.
100% open-source, 137k stars ๐
We certainly don't do everything right, but I'm proud of how much knowledge @google shares for free.
Here's an outstanding new, ungated paper from @antoniogulli and Kimberly about agent session and memory. Check it out.
https://t.co/ud3BAkoiAY
Your RAG system is failing?
It's not your vector database - it's how you're chunking your data.
Everyone focuses on picking the perfect vector database or embedding model. But here's what actually makes or breaks your RAG system: how you prepare your data before it ever gets embedded.
The fundamental challenge with chunking is this: your chunks need to be small enough for precise vector search, but complete enough to give your LLM the context it needs to generate accurate responses. Get this wrong, and even the best retrieval system will fail.
๐๐ต๐๐ป๐ธ๐ถ๐ป๐ด ๐๐๐ฟ๐ฎ๐๐ฒ๐ด๐ถ๐ฒ๐ (๐ณ๐ฟ๐ผ๐บ ๐๐ถ๐บ๐ฝ๐น๐ฒ ๐๐ผ ๐ฎ๐ฑ๐๐ฎ๐ป๐ฐ๐ฒ๐ฑ ):
๐๐ถ๐ ๐ฒ๐ฑ-๐ฆ๐ถ๐๐ฒ โ Split by token count (~512). Fast and simple, but cuts mid-sentence.
๐ฅ๐ฒ๐ฐ๐๐ฟ๐๐ถ๐๐ฒ โ Splits by structure (paragraphs โ sentences). Best for articles.
๐๐ผ๐ฐ๐๐บ๐ฒ๐ป๐-๐๐ฎ๐๐ฒ๐ฑ โ Uses inherent structure (headers, HTML tags). Perfect for structured content.
๐ฆ๐ฒ๐บ๐ฎ๐ป๐๐ถ๐ฐ โ Detects topic changes via embeddings. Variable-length chunks, one idea each.
๐๐๐ -๐๐ฎ๐๐ฒ๐ฑ โ Uses AI to identify propositions. High quality, high cost.
๐๐ด๐ฒ๐ป๐๐ถ๐ฐ โ AI agent decides which strategy to use per document. Custom but expensive.
๐๐ฎ๐๐ฒ ๐๐ต๐๐ป๐ธ๐ถ๐ป๐ด โ Embeds full document first, then derives chunks. Retains full document context.
๐๐ถ๐ฒ๐ฟ๐ฎ๐ฟ๐ฐ๐ต๐ถ๐ฐ๐ฎ๐น โ Multiple layers at different detail levels. Start broad, drill down when needed.
๐๐ฑ๐ฎ๐ฝ๐๐ถ๐๐ฒ โ Dynamic chunk sizes based on content density. Small for complex sections, larger for simple ones.
๐๐๐ ๐ต๐ฒ๐ฟ๐ฒ'๐ ๐๐ต๐ฎ๐ ๐บ๐ผ๐๐ ๐ฝ๐ฒ๐ผ๐ฝ๐น๐ฒ ๐บ๐ถ๐๐:
Chunking isn't a standalone problem. It's one piece of ๐๐ผ๐ป๐๐ฒ๐ ๐ ๐๐ป๐ด๐ถ๐ป๐ฒ๐ฒ๐ฟ๐ถ๐ป๐ด โ the architecture that determines what information your LLM sees and when.
You also need query augmentation, agents, memory, and tools working together. You can't optimize chunking in isolation.
We just released a guide covering all of this, including decision trees to help you pick the right chunking strategy for your use case.
Get your free copy here โ https://t.co/FRPyeDeyPh
First tools, then memory...
...and now there's another key layer for Agents.
Karpathy talked about it in his recent podcast.
Tools help Agents connect to the externalย world, and memory helps them remember, but they still can't learn from experience.
He said that one key gap in building Agents today is that:
"They don't have continual learning. You can't just tell them something and they'll remember it."
This isn't about storing facts in memory, but rather about building intuition.
For instance, when a human masters programming, they don't just memorize syntax.
Instead, they develop heuristics, learn edge cases, understand context, and build genuine expertise/skills through repeated interaction.
But current agents typically start from scratch every time.
Karpathy mentioned one possible path forward, which is to provide Agents with some kind of "distillation phase" that takes what happened during interactions, analyzes it, generates synthetic examples, and updates their understanding via RL.
This is similar to how humans consolidate experiences into learning.
Composio is actually building the infrastructure to solve this and provide a shared learning layer for Agents to evolve.
Think of it as the "skill layer" that gives Agents an interface to interact with over 10k tools while building practical knowledge from those interactions.
Interestingly, this direction also aligns with what Anthropic is exploring, codifying repeated agent behaviors as skills .md files.
Both approaches point toward a similar design pattern where agents progressively turn experience into reusable, composable skills.
So when one agent learns how to handle specific API edge cases, that knowledge becomes available to every other agent via Composio's collective AI learning layer, resulting in Agents that don't just automate but rather develop real intuition.
This is what Karpathy meant by continual learning, where Agents don't just memorize, but accumulate skills as they interact.
I have shared the Composio GitHub repo in the replies!
LeetCode was HARD until I Learned these 15 Patterns:
1. Prefix Sum
2. Two Pointers
3. Sliding Window
4. Fast & Slow Pointers
5. LinkedList In-place Reversal
6. Monotonic Stack
7. Top โKโ Elements
8. Overlapping Intervals
9. Modified Binary Search
10. Binary Tree Traversal
11. Depth-First Search (DFS)
12. Breadth-First Search (BFS)
13. Matrix Traversal
14. Backtracking
15. Dynamic Programming Patterns
I wrote a detailed article on these patterns and provide links to leetcode problems you can practice to learn them better.
Check it out here: https://t.co/dFv9UPKFUB
Subscribe for more such articles every week.
OpenAI Co-founder Andrej Karpathy explains the new computing paradigm:
"We're entering a new computing paradigm with large language models acting like CPUs, using tokens instead of bytes, and having a context window instead of RAM.
This is the Large Language Model OS (LMOS)"
All Companies Placement Materials๐ฅ๐ฑ
Just Free Of Cost!!
Simply:
1. Follow (So that I can DM)
2. Like and Repost
3. Comment "placement" to recieve your copies