If you had to pick one, what "AI Task" is most production-ready?

🤖 A new version of NeuML's RAG app is available! This version integrates the latest from txtai and updates the default LLMs. github.com/neuml/rag

😎 With AI Agents you'll quickly realize that you like determinism. LLMs don't always go down the same path. People jump to AI Agents because it's what all the cool kids are doing. But what many need are workflows. Workflows can chain functions, LLM calls or other transformers

⭐ What's going to happen to the AI space when someone releases a new architecture that is intelligent, trains with a little data and is small enough to run only on CPUs? The whole ecosystem is propped up by the concept that you need massive amounts of hardware.

🚀 A TxtAI Agent to write a paper about TxtAI? Have to say this is quite amazing! Check out this example that prompts an agent to research TxtAI and then write an Arxiv-style research paper. All with an open 4B parameter model. Code: gist.github.com/davidmezzetti/… Generated Paper:

AI-guided coding (vibe coding is a terrible phrase that leads you to believe it's not serious) shouldn't be a culture war. If your brain can come up with it faster, do it. If asking an LLM is faster, do it. I don't see what all the conversation is about.

🧬⚕️🔬 If NeuML had to be pinned to one vertical, it would be medical research. Check out this notebook that covers building a RAG pipeline for PubMed documents. github.com/neuml/txtai/bl…

Licensing with AI models is so nebulous. Some are willing to release with permissive licenses designed for software (MIT, Apache) which is great for adoption. But it's still unclear how it applies to AI models. Others make up their own licenses which is risky for businesses.

Did you know that PaperETL can generate a citation graph using the PubMed baseline? Check out this dataset of the Top 100 most highly cited PubMed articles. Interesting to see a mix of DNA sequencing, cancer research and of course COVID-19 articles. huggingface.co/datasets/NeuML…

RAG != Vector Search. While often paired together they don't have to be. Any of the following are OK. Websearch RAG SQL Query RAG Keyword Search RAG Bash script RAG

Great to see that someone applied what we did with BERT Hash, ColBERT and MUVERA to Turkish models! The power⚡ of open source at work! Link to ArXiv paper: arxiv.org/abs/2511.16528 huggingface.co/blog/nmmursit/…

📄 ⚙️ If you're in the medical space, you should check out PaperETL. PaperETL can process a number of medical literature formats including the PubMed baseline. Subsets of PubMed can be built using a list of ids or series of MeSH codes. Once created, PaperETL databases can be

🚀 GraphRAG is a popular concept but what is it? TxtAI was one of the first to the scene with GraphRAG in 2022. It utilizes a vector index to automatically construct a graph network of nodes between each of the indexed records. This enables a different type of similarity query.

⚡ Training a text classifier with LLM-generated data is more powerful than it appears. The text classifier has the advantage of seeing the entire training set and learning all of the connections. Applying a LLM prompt to a single record is only looking at that one instance.

⭐ Interested in Astronomy? Then check out this TxtAI example that extracts constellation data from Wikipedia and builds a knowledge graph connecting the stars! github.com/neuml/txtai/bl…