Malte Pietsch

440 posts

Malte Pietsch

Malte Pietsch

@malte_pietsch

Co-Founder & CTO of deepset | Building Haystack & deepset Cloud | NLP, LLMs, Engineering, Open-source

Berlin, Deutschland Katılım Nisan 2010
219 Takip Edilen492 Takipçiler
Malte Pietsch
Malte Pietsch@malte_pietsch·
interesting analysis @hwchase17 - though you've forgotten your friends at Haystack 😉 We seem to share the view that blending "AI workflows" with "agentic" capabilities gives best results right now, and we extended Haystack with many carefully chosen agentic capabilities over the last year. As the thread here indicates, there are plenty of directions agent frameworks can focus and evolve – how about we get together with @jerryjliu0 @samuelcolvin @MarcKlingen on a live panel to discuss? Would for sure be a panel with many hot takes and those are typically the most interesting ones 😅
English
0
5
18
614
Harrison Chase
Harrison Chase@hwchase17·
OpenAI recently released a guide on building agents which contains some misguided takes There's a lot of FUD, confusion, hype, and noise around agents I wrote a blog on how to think about agent frameworks. Includes: Background Info - What is an agent? - What is hard about building agents? - What is LangGraph? Flavors of agentic frameworks - “Agents” vs “workflows” - Declarative vs non-declarative - Agent abstractions - Multi agent Common Questions - What is the value of a framework? - As the models get better, will everything become agents instead of workflows? - What did OpenAI get wrong in their take? - How do all the agent frameworks compare?
Harrison Chase tweet media
English
88
120
920
450K
Malte Pietsch retweetledi
Haystack
Haystack@Haystack_AI·
🚀 Haystack 2.10.0 is here – packed with powerful upgrades for agentic pipelines!⁣ ⁣ - Smarter 𝐏𝐢𝐩𝐞𝐥𝐢𝐧𝐞.𝐫𝐮𝐧() logic fixes execution issues and improves the reliability of complex pipelines⁣ - 𝐀𝐬𝐲𝐧𝐜𝐏𝐢𝐩𝐞𝐥𝐢𝐧𝐞 enables huge improvements through concurrent component execution: 𝟏.𝟕𝟓𝐱 𝐬𝐩𝐞𝐞𝐝 in hybrid retrieval and 𝟑𝐱 𝐬𝐩𝐞𝐞𝐝 in parallel generation tasks - 𝐓𝐨𝐨𝐥 𝐜𝐚𝐥𝐥𝐢𝐧𝐠 𝐞𝐯𝐞𝐫𝐲𝐰𝐡𝐞𝐫𝐞, making it easy to switch between chat generators - 𝐋𝐨𝐜𝐚𝐥 𝐩𝐢𝐩𝐞𝐥𝐢𝐧𝐞 𝐯𝐢𝐬𝐮𝐚𝐥𝐢𝐳𝐚𝐭𝐢𝐨𝐧, render graphs without internet - 𝐒𝐦𝐚𝐫𝐭𝐞𝐫 𝐝𝐨𝐜𝐮𝐦𝐞𝐧𝐭 𝐩𝐫𝐨𝐜𝐞𝐬𝐬𝐢𝐧𝐠, with CSV splitting, cleaning & LLM-powered metadata extraction ✨ Want to deploy your pipeline easily? Use 𝐇𝐚𝐲𝐡𝐨𝐨𝐤𝐬 to serve Haystack pipelines as REST APIs! ⁣ Check out the full release notes for more details 👇
Haystack tweet media
English
2
4
11
1.1K
Malte Pietsch retweetledi
GV
GV@GVteam·
Congratulations to @deepset_ai on being recognized in the 2024 Gartner® Cool Vendors in AI Engineering report! We’re excited to support @rusic_milos, @malte_pietsch, and the team as they simplify the complexity of compound AI development.
deepset, makers of Haystack@deepset_ai

Excited to share that deepset has been recognized as a 2024 Gartner® Cool Vendor in AI Engineering. We believe this recognition validates our innovative AI approach and work empowering enterprise AI adoption. Get the report: hubs.li/Q02YfTv50

English
1
3
6
5.8K
Malte Pietsch retweetledi
NVIDIA AI Developer
NVIDIA AI Developer@NVIDIAAIDev·
✅Get ready to take your AI RAG and agentic projects to the next level. Simplify and accelerate building simple to complex #AI applications with Haystack, whether you're working in the cloud or on-prem. The integration of NVIDIA AI Enterprise with @deepset_ai Studio is here 👇
deepset, makers of Haystack@deepset_ai

🚀 Live from #Ai42024! Thrilled to announce deepset Studio, an interactive tool for designing custom AI pipelines with @Haystack_AI! With YAML export and integrations to deepset Cloud & @NVIDIA AI Enterprise, you can deploy mission-critical apps more efficiently than ever. Learn more: bit.ly/deepset-studio… #AI #deepset #haystack #NVIDIA #LLM

English
1
8
54
3.6K
Malte Pietsch
Malte Pietsch@malte_pietsch·
Interested in latest embedding models? We share insights from our latest work together with @mixedbreadai where we trained a SOTA German embedding model...
Haystack@Haystack_AI

🚨 Upcoming livestream alert 🚨 Join @tuanacelik from @deepset_ai and @aaxsh18 from @mixedbreadai on August 7th to learn about our jointly released embedding model: deepset-mxbai-embed-de-large-v1 We will discuss binary quantization, MRL, and how you can start using this model with Haystack 🚀 Register here and join us on YouTube 👇 lu.ma/4bzy3d5q

English
0
1
4
91
Malte Pietsch retweetledi
Tuana
Tuana@tuanacelik·
New projects for @Haystack_AI and the AI community coming soon! Thank you to @AndrewYNg and @DeepLearningAI for the hospitality last week. And I’m looking forward to sharing our work with everyone 🎸
Tuana tweet mediaTuana tweet media
English
5
9
88
5.4K
Malte Pietsch retweetledi
Tuana
Tuana@tuanacelik·
3 days ago @qdrant_engine announced a new baseline for hybrid search: BM42 ​ My understanding is that the work stems from the observation that typical search algorithms such as TF-IDF and BM25 (which evolved from TF-IDF) are powerful, but not the best suited for RAG applications. Where documents are usually shorter. ​ BM42 takes a similar approach to SPLADE to resolve this issue, but differently to SPLADE, BM42 is based on lightweight sentence-transformers models, with no need for fine-tuning. BM42 is also computationally lighter than SPLADE. ​ You can try out this approach in a hybrid retrieval pipeline in the latest cookbook by @theanakin87 👇 github.com/deepset-ai/hay… ​ Or read about BM42 in more depth here: qdrant.tech/articles/bm42/
Tuana tweet media
English
1
20
77
7.3K
Malte Pietsch retweetledi
Vladimir Blagojevic
Vladimir Blagojevic@vladblagoje·
No need to rely solely on closed models like gpt-4 for evals. Prometheus 2, an open-source alternative, offers high correlation with human evaluations, flexibility, transparency, and without the downsides of closed models (privacy, cost etc). Check out this blog post and a Colab by my teammate @theanakin87
Haystack@Haystack_AI

Prometheus 2 is a SOTA evaluation model, fine-tuned from models by @MistralAI, specifically to evaluate responses generated from LLMs. ​ This model uses a special prompt, describing exactly how you would like your responses to be evaluated. ​ You can learn all about exactly how to use Prometheus, and how you can evaluate your RAG pipelines with it in this article by @theanakin87 ​ haystack.deepset.ai/blog/rag-evalu…

English
0
1
9
425
Malte Pietsch retweetledi
Haystack
Haystack@Haystack_AI·
Want to challenge Llama 3? 🦙🦙🦙 ​ Introducing 🧑‍🏫 AutoQuizzer, a fun and educational LLM quiz generator built with 🦙 Llama 3 8B Instruct,⚡ @GroqInc, and Haystack! 💙 ​ 🤹 How It Works: • Provide a URL -> Instantly get a multiple-choice quiz. • Play it yourself or let the LLM take over in two modes: 📕 Closed book: Uses its own knowledge. 🔎 🌐 Web RAG: Searches Google for top snippets to answer. ​ Give it a try on @huggingface Spaces 🤗 huggingface.co/spaces/deepset…
Haystack tweet media
English
0
3
10
2.3K
Malte Pietsch retweetledi
Tuana
Tuana@tuanacelik·
If you have NLP pipelines running in production, you likely need to maintain them. If you need to detect the sources of problems, having a good way to see each step of execution, and even more, identifying the pipeline with issues (when you have more than one running) is quite important. ✨ Queue the tracing support we introduced in @Haystack_AI 2.0 ✨ Out of the box, Haystack supports @opentelemetry and @datadoghq But we've also added a uniform Tracer interface that allows you to extend Haystack to support other platforms too 🙌 See how to connect your own tracers here 👉 #custom-tracing-backend" target="_blank" rel="nofollow noopener">docs.haystack.deepset.ai/docs/tracing#c… We've also included instructions on how you can set up a lightweight local tracing backend with a visualizer using @JaegerTracing in our guide too 🚀
Tuana tweet media
English
0
5
11
1.5K
Malte Pietsch retweetledi
Haystack
Haystack@Haystack_AI·
We are so happy with the reception to Haystack 2.0's launch! If you want to write production-ready Python NLP apps with a stable, transparent, customizable framework, give it a try here: haystack.deepset.ai/overview/quick… Our open-source community are the real heroes here - we can’t thank you enough. Over the past few years of rapid change in the LLM universe, our users, testers, and contributors have taken the time to give us invaluable feedback about their needs. Rewriting is a big decision; community involvement has given us the confidence that we’re moving in the right direction. 2.0 is a better foundation that will serve developers’ needs for years to come. If you haven’t tried out Haystack 2.0, there are loads of materials to get you started: 📖 Documentation: docs.haystack.deepset.ai/docs/intro 📘 Tutorials: haystack.deepset.ai/tutorials 🧑‍🍳 Cookbooks: github.com/deepset-ai/hay… 📚 Blog Posts: haystack.deepset.ai/blog Happy building! #HaystackV2 #OpenSource #LLM #python
Haystack tweet media
English
0
7
11
1.3K
Malte Pietsch retweetledi
Haystack
Haystack@Haystack_AI·
We are thrilled to announce the stable version of Haystack 2.0 🎉 We’ve been working on this for a while and now Haystack 2.0 has everything to help you implement composable LLM applications that are easy to use, customize, extend, optimize, evaluate, and deploy to production. We are hosting an office hour session about Haystack 2.0 tomorrow in the Haystack Discord 🧡 Drop by to ask anything about this release and Haystack 👉 discord.gg/WTXcUW6w?event… Learn more about the release 👉 haystack.deepset.ai/blog/haystack-… Give Haystack 2.0 a try 👉 haystack.deepset.ai/overview/quick… Check out all features in our release notes 👉 haystack.deepset.ai/release-notes/… #HaystackV2 #LLM #opensource #Python
Haystack tweet media
English
2
27
78
20.2K
Malte Pietsch retweetledi
Haystack
Haystack@Haystack_AI·
Two weeks to go for the Haystack 🧡 @neo4j webinar 📆 March 21st, 6PM CET (9AM PT) ​Talk 1: Neo4j Haystack Integration with Sergey Bondarenco Talk 2: Knowledge Graphs for RAG with @akollegger Sign up here: lu.ma/6iskyd0g
English
0
4
8
2K
Malte Pietsch retweetledi
Haystack
Haystack@Haystack_AI·
We're adding guides on Advanced RAG Techniques. Our first one: HyDE (Hypothetical Document Embeddings): Read the full guide to learn: · When HyDE is useful · How it works · How To Build It in Haystack 👉 docs.haystack.deepset.ai/v2.0/docs/hypo…
Haystack tweet media
English
0
6
6
663

Keşfet

@hwchase17 @jerryjliu0 @samuelcolvin @MarcKlingen @deepset_ai @rusic_milos @Haystack_AI @NVIDIA