Himanshu Verma

The future of GraphRAG isn't a bigger graph. It's a graph that can admit: "I don't know." Then go find the answer, rewrite itself, and continue reasoning.

Wow, @milichab texted me five days ago, and I just saw this.

@saen_dev @DailyDoseOfDS_ I've been contributing to HugeGraph-AI, and lately I've been reading a lot of papers on graph construction. RAGU was a particularly interesting read.

@Bitflicker64 @DailyDoseOfDS_ @Bitflicker64 Followed back! The pace right now is genuinely insane — what part of the space are you building or watching most closely?

RAG vs. Graph RAG vs. Agentic RAG, clearly explained! Standard RAG embeds documents into vectors and retrieves the most similar chunks via similarity search. For direct factual lookups, this works well. But it breaks down when a query needs to connect facts spread across multiple documents. Similarity search retrieves individual chunks, not the relationships between them. Graph RAG adds a knowledge graph layer on top. → During indexing, an LLM extracts entities and relationships from the documents. → During retrieval, the system traverses these connections instead of relying on embedding similarity alone. This is what enables multi-hop queries. Say a vector DB stores three facts about internal services: ↳ "The checkout service uses payments API." ↳ "The payments API runs on cluster-3." ↳ "Cluster-3 is scheduled for maintenance on Friday." Someone asks: "Will the checkout service be affected by Friday's maintenance?" Vector search can likely retrieve facts 1 and 3 because the query mentions "checkout service" and "Friday maintenance." But it will miss fact 2, which connects the payments API to cluster-3. That middle fact sits too far from the query in embedding space. It mentions neither "checkout" nor "maintenance," so it never makes it into the retrieved context. A knowledge graph connects these as linked entities, and graph traversal finds the full path in one query. Agentic RAG takes a different approach entirely. Instead of a fixed retrieval pipeline, an LLM agent decides at query time which tools to invoke, which sources to query, and in what order. Check the visual below to understand the three architectures thoroughly. One thing to note here is that these three aren't levels of sophistication that you need to graduate through. Instead, they solve different query types. ↳ Single-hop factual lookups → standard RAG ↳ Multi-hop relationship queries → Graph RAG ↳ Dynamic multi-source tasks with tool use → Agentic RAG Once the right architecture is in place, the next leverage point is efficiency. Most RAG architectures rely heavily on vector search, and that layer can be made 32x more memory efficient using binary quantization. We covered the full implementation in the article below. 👉 Over to you: Which RAG architecture are you running in production?

@AlexFinn Apple might win the local AI game, but I’m more curious if developers will actually make apps that take advantage of that 1TB memory, or if everything just becomes a fancier local ChatGPT

Apple will win the AI war in June 2026 without investing a single penny into AI The war won without a sword They will release the M5 Ultra with 1TB of memory It will be able to run super intelligence more powerful then Opus locally on your desk for free It will fundamentally change the way the entire world operates. EVERYONE with a personalized super intelligent AI employee working on their desk 24/7 $200 a month AI plans will feel antiquated While you've been crapping on Apple for not entering an AI race against 100 other companies shredding trillions of dollars, they ran their own race Personalized local super intelligence will be stage 1 of the singularity. I sincerely hope you're preparing for this

@saen_dev @DailyDoseOfDS_ IKK RYTT but i think this area is developing rapidly too !! wanna connect ? i followed you

@Bitflicker64 @DailyDoseOfDS_ Multi-stage extraction with entity validation producing cleaner graphs than single pass is exactly what the research keeps confirming. Higher evidence recall and more faithful answers are the outcome everyone wants but few invest the engineering effort to get there.

Billions of parameters. Finite resources.

Of course, I wanted to. Took forever, but now when I open the graph I can actually see how everything connects instead of just staring at a mess. Worth it.

"Your graph is cute."~ This is what happens when you actually link your shit instead of hoarding notes like a digital hoarder. If yours doesn’t look this connected, don’t even bother replying.

@mdkaifansari04 Gave me a generic line to Post Nthg good

@Bitflicker64 let me know what it come up with

@DREAMSTICK4 Ahahhahah

@Bitflicker64 Just 12% bro😭

Be honest... what % were y'all at before they reset the limits? 😭 I only got to 26%

When I realized nobody knows what they are doing, it's idiots all the way up. You don't need qualifications or permissions or expertise. You can literally just do things.

@saen_dev @DailyDoseOfDS_ I've been reading quite a bit about this recently, and it's interesting how many newer approaches are moving toward multi-stage extraction with validation. I haven't implemented it myself yet, but it definitely seems like a more robust direction than single-pass extraction.

@Bitflicker64 @DailyDoseOfDS_ Multi-stage extraction with entity validation is the approach I keep coming back to as well. Single pass graphs end up with tons of duplicate and conflicting nodes. Are you running validation as a separate LLM call or using a rule based system for entity dedup?

YOOOOOO!!!!! theyyy did itt????!!!!!!!

I used the new GPT - 5.6, and the review was excellent. When I reviewed a PR, it received a thorough human review. !!!

GraphRAG isn't the bottleneck anymore.... graph quality is. The latest GraphRAG research is increasingly focused on extraction pipelines and graph construction rather than retrieval algorithms themselves.

@saen_dev @DailyDoseOfDS_ Building the graph is only half the problem. Recent research shows multi-stage extraction with entity validation produces much cleaner graphs than single-pass extraction, leading to higher evidence recall and more faithful answers.

@DailyDoseOfDS_ Graph RAG solves multi-hop problems but your knowledge graph quality becomes the new bottleneck. Most teams spend months building the graph only to realize entity extraction was wrong from the start. How do you validate graph quality at scale?

@0xrinx real

I need to work more

@ericciarla bruh u hire like all the time do u even have so many positions open 🤧🤧🤧🤧🤧

we're hiring