Juan David Gómez

Full benchmark with architecture diagrams, code, and results (open source): dev.to/juandastic/i-b…

You can't ask "find the most connected entity and give me all its facts" if the facts live in a different store than the structure. I'm calling this "context blindness" and for a personal companion it matters more than retrieval scores.

I benchmarked @zep_ai vs @mem0ai to find out if I was over-engineering my AI companion's memory. Both solve fact extraction, deduplication, and contradiction handling. But they store data in fundamentally different ways.

@dfect I think it is actually very clever. The website is not the product you will use daily; it should not be usable, it has to be memorable, and it definitely achieves that

Creating an account with Posthog for the first time, the website made me feel all kind of different feelings. A) It's too many options and things B) I like it, it's geeky C) It's funny I don't know if i like it or I hate it.

@BrendanFalk I am not looking for a new job, but the idea is so good that it makes me want to apply just to challenge myself with this approach

I believe we've found the best AI-native coding interview We call it the “Composer 1 interview” Candidates get 1 hour to build a real, medium-sized project live The only constraint: they have to use Cursor’s Composer 1 model

@VexrailHQ Building for AI is the new "luxury" side project. Back in the day, you could launch a cool idea for free and just let it live. Now, with AI costs, monetization isn't a choice; it’s a survival tactic. This is a clever way to keep projects sustainable without breaking the bank.

Check out our latest post: vexrail.com/resources/blog…

I built a bidirectional sync between a Neo4j Knowledge Graph and Notion for the @ThePracticalDev Challenge with @NotionHQ AI memory is usually a black box. Now, it's a fully editable, dynamically generated Notion workspace. Here is the architecture dev.to/juandastic/ful…

@meta_alchemist I had a chance to use Graphiti for my side project and it is great and really useful, could be expensive since turning text into graph consume tokens but for the rigth problem it make sense and it is worth. I post my experience in case someone is interested

AI normally sucks at memory retrieval both you and your AI hate that Here are the top 10 open source AI memory layers > free > open source > with lots of GitHub stars > and some even funded by YC You can use them to make your agents, claude, codex, and openclaw much smarter Also, tips on what each memory is good at, how to combine them, and get even more out of them are all in the article. These are much better than just memory md files, and will make a huge difference in your workflow! Give your agents life with them:

@ThePracticalDev The magic of building a product that someone uses is watching how your code evolves and behaves over time and taking care of it like a garden. Software is not a static thing; it is alive, and it evolves with your users. This article is just an example of how it looks in real life

Context windows are expensive resources. This dev demonstrates how to tame a 120k token prompt by implementing a deterministic GraphRAG approach for more efficient memory scaling. { author: @juandastic } dev.to/juandastic/sca…

The Result: Token usage collapsed back to my 40k limit. Latency stayed under 1s. The AI still feels like it knows everything. I wrote a full deep dive on @ThePracticalDev with the architecture dev.to/juandastic/sca…

To get those details back, I built Deterministic GraphRAG. I hate Agent tool-calling (adds 3-5s latency). Instead, a straight-line Hybrid Search runs in <1s. It checks what's already in the prompt and only injects missing facts. Zero redundancy.

1 Million token context windows are a luxury, not an architecture. My AI side-project hit a wall when the system prompt bloated to 120,000 tokens per message. Here is how I fixed it using a Knowledge Graph, a "Waterfill" budget, and Agentless GraphRAG. 🧵👇

@the_shujaa @convex Congratulations on the new sales. You indeed have a robust proposal for memory, and the convex integration makes it more interesting

@convex I have not even posted on any platform yet except a reply I made here on twitter which got zero likes or bookmarks. I guess time to start advertising now.

😮😮Is this normal made two Teams sale on memcity.dev a Rag and memory component running entirely on @convex I have done zero marketing whatsoever am just scared broo🙆‍♂️🙆‍♂️. If you have made a purchase DM I help you with integration and any bug fixes.

@waynesutton @the_shujaa Great project, I was also experimenting with RagGraphs and Convex, definitely a powerful combination for a great UX and robust Memory solution

memcity by @the_shujaa AI Memory for Convex Apps. Vector search, knowledge graph, episodic memory, and a production-grade 16-step RAG pipeline. memcity.dev

A look at what people are building with @convex this week.

Congrats to our top 7 authors this week! 🏆 @dannwaneri, @madsstoumann, Julien Avezou, Peter Mulligan, Prithwish Nath, Vivek V, and @juandastic. This week's lineup features a CSS recreation of a Pantone color deck, a self-hosted Google Trends alternative using DuckDB, and much more. dev.to/devteam/top-7-…

Hey @GustyCube, I just saw your project on a @theo stream. I am also interested in the same problem. I experimented with Graphs, aiming to solve the same deep memory problem. I posted something about it in case you are interested. I will definitely go check Membrane to know more