TNG Technology Consulting GmbH

Welcome to Big Techday 25! Motorworld Munich is slowly filling up. Our program has a clear focus on #AI, but also covers topics such as #robotics, software engineering, hardware hacking, and adventure. You can now find the links to the livestreams on our website: www.bigtechday.com/en/agenda #tngbtd

Last week, our #Atlassian Consulting Team attended Atlassian Team ’25 Europe in Barcelona. We showcased our apps [1] for #Jira and #Confluence at a booth and discussed new ideas with customers, partners, and other vendors.

[1] marketplace.atlassian.com/vendors/114/...

We recently integrated Real-Time Action Chunking (RTC) into an Nvidia ISAAC-GR00T N1.5 Robot Foundation Model. It enables our SO101 robot arm to move much more smoothly and natural. Check out the comparison between our fine-tune and the original baseline model.

Compact #computing power at TNG: We've created a small computer equipped with an RTX 6000 Pro and 96 GB VRAM. It’s ideal for hosting large #LLMs and other #AI models. We intend to use it for local benchmark tests and showcases.

TNG joined e-fellows.net’s event Women in Technology in Munich where female #STEM students connected with renowned #tech companies. Our colleagues held one-on-ones and presented their talk “Business Analytics: The intersection of business and IT”. 💡

At the #IFA2025 in Berlin, our client BSH presented three use cases co-developed with TNG: a chatbot for controlling household appliances, a partnership for electricity price-optimized device control, and an #AI recipe converter for the #Cookit. From minute 9:20: www.tagesschau.de/tagesschau_2...

Last weekend, we hosted #MuniHac 2025. 70 #Haskell fans from around the world enjoyed a packed schedule of coding sessions, hacking projects, and talks, including keynotes by Andrew Lelechenko, Mike Sperber, and Rodrigo Mesquita.

Find the recordings of the keynotes here: youtube.com/playlist?lis...

#𝗧𝗡𝗚𝗔𝗜𝗜𝗻𝘀𝗶𝗴𝗵𝘁 𝗤𝘄𝗲𝗻 𝟯 𝗖𝗼𝗱𝗲𝗿 𝟰𝟴𝟬𝗕 + 𝗤𝘄𝗲𝗻 𝗖𝗼𝗱𝗲 𝗖𝗟𝗜
A state-of-the-art Open Source model for #agenticcoding alongside a CLI-based tool. Self-hosting this model allows usage in projects that require strict data privacy without third-party dependencies. More information: www.tngtech.com/en/about-us/...

Migrating a 600-module Java codebase: How TNG tackled a complex enterprise upgrade Authors: Michael Pisula & Dr. Philipp Schuegraf At TNG, we like to solve hard IT problems.

𝗠𝗶𝗴𝗿𝗮𝘁𝗶𝗻𝗴 𝗰𝗼𝗺𝗽𝗹𝗲𝘅 𝗲𝗻𝘁𝗲𝗿𝗽𝗿𝗶𝘀𝗲 𝗝𝗮𝘃𝗮 𝗰𝗼𝗱𝗲 𝘄𝗶𝘁𝗵 𝗔𝗜 𝗮𝘀𝘀𝗶𝘀𝘁𝗮𝗻𝗰𝗲
We migrated a 1.5M-line legacy #Java codebase to a modern stack using our Java AI Migration Tool (JAMI), achieving an average #migration speed of 30 minutes per module. Read about our approach and learnings: www.linkedin.com/pulse/migrat...

We have developed 𝗧𝗡𝗚 𝗔𝗜 𝗘𝟮𝗘 𝗧𝗲𝘀𝘁𝗚𝗲𝗻, an AI tool for automated and efficient end-to-end test generation for web applications. It generates clean, maintainable Playwright tests in Python, Java, C#, and TypeScript without requiring access to the source code. Check out: www.tngtech.com/en/about-us/...

🚀 Table Enhancer for #Confluence just got smoother. With our new #tngapp feature, you can drop your table directly inside the macro - no more extra steps or page hopping. Simple, fast, and seamless🪄 Check out at #atlassian 👉https://marketplace.atlassian.com/apps/1211609/table-enhancer-for-confluence

AI-assisted PL/I to Java Migration Authors: Andreas Bärnthaler, Dr. Eric Schlarmann, and Mathieu Desponds Motivation In this article, we show you how TNG Technology Consulting uses Artificial Intelligence (AI), and in particular Large ...

Our article "AI-assisted PL/I to Java Migration" explores how legacy code can be modernized and migrated to #Java using #AI accurately and fast. Learn more about our innovative approach on #CodeMigration which combines #LLMs with rule-based techniques 👉 www.linkedin.com/pulse/ai-ass...

#𝗧𝗡𝗚𝗔𝗹𝗹𝗻𝘀𝗶𝗴𝗵𝘁 𝗠𝗖𝗣 – 𝗠𝗼𝗱𝗲𝗹 𝗖𝗼𝗻𝘁𝗲𝘅𝘁 𝗣𝗿𝗼𝘁𝗼𝗰𝗼𝗹
Standardizes how applications and services exchange structured information in a reliable, predictable way.
Benefits: Better interoperability, modularity & scalability.
More: modelcontextprotocol.io/docs/getting...

Tired of manually translating Confluence pages?
With our #tngapp Next Generation #Translator for #Confluence you can now enable our new feature ✨ - Auto-translation. Pages in any language are instantly translated to your preferred one. Check out #atlassian 👉 marketplace.atlassian.com/apps/1225038...

We recently presented a new prototype for urban planning at the Open Hertzlab of the ZKM Center for Art and Media. Visitors re-designed Karlsruhe’s Market Square using an image AI that provided direct feedback on how the landscape would change, allowing users to imagine the future of urban living.

We have confirmed three keynotes for MuniHac 2025. Join us in Munich between September 12th and 14th for talks by Andrew Lelechenko, @sperbsen.bsky.social, and Rodrigo Mesquita as well as workshops and hacking projects: munihac.de/2025.html#re...

#MuniHac #Haskell

We once again participated in the #B2Run at #Munich. The event offered the chance to engage in #sports activities and bond with teammates. We are already looking forward to the upcoming company run events and wish lots of fun and success. 👟

Save the Date: Big Techday 25 will take place at Motorworld Munich on October 24th. Our about 1,250 guests will enjoy talks by 40+ speakers, with a focus on #AI topics. More information on our conference: www.bigtechday.com/en/

Our recent summer retreat brought us together at seven TNG offices across Europe. We talked about strategic challenges, the implications of AI for our daily work, and how to improve our internal processes. With these discussions, we want to benefit both our clients and colleagues.

On July 18th, we will present an innovative prototype for AI-supported urban development in cooperation with the Center for Art and Media (ZKM) #Karlsruhe. Visitors can use an interactive image #AI to virtually design a selected location in Karlsruhe. More information 👉 zkm.de/en/2025/07/o...

One of our colleagues created a three-part article series on the topic of #eventmodeling. Part II of the article series examines the claims of Event Modeling, arguing it's effective for specific domains, but not a one-size-fits-all solution. Read the article here 👉 www.linkedin.com/pulse/event-...

Lünendonk-Liste 2025: Mittelständische IT-Beratungen zu abhängig von Automobilbranche und Softwareentwicklung - Luenendonk Die korrespondierende Lünendonk-Liste zur Studie "Der Markt für IT-Dienstleistungen in Deutschland" - jetzt kostenfrei downloaden!

TNG secures 14th place in the Lünendonk® List 2025 of leading mid-sized IT consulting and systems integration firms in Germany. Further information can be found here:
www.luenendonk.de/luenendonk-l...

tngtech/DeepSeek-TNG-R1T2-Chimera · Hugging Face We’re on a journey to advance and democratize artificial intelligence through open source and open science.

The model weights are available on Hugging Face under the MIT licence: huggingface.co/tngtech/Deep...

Today, we release DeepSeek-TNG R1T2 Chimera, a Tri-Mind Assembly-of-Experts LLM constructed using DeepSeek's R1-0528, R1 and V3-0324. It operates at a sweet spot in intelligence vs. inference cost and appears to be smarter and/or faster than its parent models as well as <think>-token consistent.

Event Modeling - Part I: What it is and how it works Software development is inherently complex, involving a myriad of challenges from technology choices to design and architecture. Beyond technical aspects, it requires a deep understanding of the domai...

One of our colleagues created a three-part article series on the topic of #EventModeling, covering the concept and its application in everyday #SoftwareDevelopment. Part I of the series outlines the core concepts of Event Modeling. Read the article here 👉 www.linkedin.com/pulse/event-...

In our first #TNGAIInsight, we introduce you to World Foundational Models (WFMs), large-scale generative AI systems designed to comprehend physics and spatial relationships.

Find out how WFMs accelerate AI development and how we plan to use these models: www.tngtech.com/en/about-us/...

Assembly of Experts: Linear-time construction of the Chimera LLM variants with emergent and adaptable behaviors Requiring $10^{13}$-$10^{15}$ FLOPs to calculate one 8 bit weight in an LLM during pretraining is extremely expensive and seems inefficient. To better leverage the huge investments made into pretraine...

We recently published a new research paper on arXiv [1] and Hugging Face [2]. It explains how we constructed the 671B R1T Chimera child model from the great #DeepSeek AI V3-0324 and R1 parent models in less than one hour of CPU time.

[1] arxiv.org/abs/2506.14794

[2] huggingface.co/tngtech/Deep...

From September 12th to 14th, we host MuniHac, an annual #Haskell community gathering in Munich. Join Haskell fans from all over the world and polish your skills in workshops, talks, and coding sessions.

More information, the registration, and our call for contributions: munihac.de/2025.html

We have developed UnitTestGen, a specialized tool for automated & efficient #unittest generation. It's designed with a strong focus on #dataprivacy and can be executed with a single command. We provide the tool directly and support its integration. More information 👉 www.tngtech.com/en/about-us/...

Last week, we got a geek boost with Matt Parker & team. He's an author, comedian, and #math guru, making complex concepts fun. They're targeting the #worldrecord in #computing π by hand. After sharing insights, we had a full-day π workshop with 30+ TNG colleagues. Thanks for the nerdy day 🤓