9.3 C
New York
Tuesday, March 5, 2024

AI Planet 2023 Recap

AI Planet 2023 Recap

AI Planet

As we stand on the threshold of bidding farewell to 2023 and eagerly anticipate the dawn of a new year, it is with great enthusiasm and pride that we reflect upon the incredible journey that has defined AI Planet over the past twelve months.

In 2023, we released our most anticipated Open Source project — GenAI Stack, an end-to-end framework designed to reliable build large language models (LLMs) based applications seamlessly.

  • Stable and Componentized Architecture: We deployed a stable, componentized architecture, ensuring better visibility and reliable execution for LLM applications. Addressing a critical industry challenge, this architecture moves beyond the common limitation where most LLM applications remain confined to prototype stages and struggle to reach production.
  • Expanded Data Connector Integration: Our ETL pipeline now includes vast array of data connectors, thanks to several opensource data frameworks including Langchain, Llamaindex, and Airbyte.
  • Built diverse LLM apps, robust RAG and Knowledge graph systems: Over time, we have developed various LLM based applications across Financial, Healthcare, Education and other industries, employing comprehensive knowledge graph and end-to-end Retrieval Augmented Generation (RAG) pipelines.

While we’re still building GenAI Stack in a closed environment, parts of our code is made open source, and we’re looking forward to releasing newer versions soon. Please feel free to contribute: https://github.com/aiplanethub/genai-stack. Don’t forget to ⭐ the repo.

AI Marketplace is a comprehensive solution tailored for enterprises to seamlessly integrate AI models into their products with secure API end points. Here are some of the key highlights and advancements we’ve achieved:

  • Robust LLM Inference Engine with Serverless: Our upgraded inference engine can deploy Large Language Models (LLMs) at scale, with enhanced throughput and latency while also optimizing costs. In addition, we’ve enabled custom CUDA versioning system and serverless inferencing with reduced cold starts, which is a cherry on the cake.
  • Diverse Range of Advanced LLMs: Our marketplace hosts 50+ advanced opensource and proprietary models, including Zephyr, Llama, the Effi series, Falcon, Text to Music, Indic Translation, and more.
  • Robust Queuing System for Improved Reliability: We have implemented a robust queuing system to effectively manage request throughput.

Each of these developments contributes to a more dynamic, efficient, and user-friendly AI integration experience for our enterprise clients.

Large Language models are all over the place. With the growing popularity of open-source models on generic data, the need for reasoning models to evaluate the LLMs comes in handy. We built our in-house Open Source Large Language Models Series with over 60,000+ overall downloads in combination.

Open Source LLMs

  • Effi series: In July this year, we released, effi 7b and effi 13b. While effi 7b, trained on instructions and based on the Llama 7b architecture, pushes the boundaries, surpassing many noteworthy models such as Llama-7b, MPT-7b and Falcon-7b. On the other hand, Effi 13B, with its unique training on a chain of thoughts, boasts heightened reasoning capabilities. Our Effi-13B model has had around 40,000+ downloads this year. We further experimented with Quantization and released the Effi-13B-Quant model that reduce inference costs by 5x without compromising quality. Read more about benchmarks: Effi LLM Series
  • Panda Coder-13B: Panda-Coder is a state-of-the-art LLM, a fine-tuned model, specifically designed to generate code based on natural language instructions. It’s the result of relentless innovation and meticulous fine-tuning, all to make coding easier and more accessible for everyone. Panda Coder has over 15000+ downloads and it can help you generate comments and documentation to your code. Learn more: Panda Coder

Closed Source LLMs & Research

  • Our contribution just doesn’t end there, we also built several closed-source domain-specific Large Language models on private data in the domains of Finance, Insurance and Healthcare.
  • Further, we also built a Hallucination-free RAG and Knowledge pipeline that can chat with documents seamlessly by using opensource LLMs.

AI Planet’s vision is to make AI accessible to everyone. And we believe creating awareness about AI and preparing the world for the AI era by taking the community along is crucial in our journey. Here is our key-highlights this year:

  • Learning.MachineLearning is now part of AI Planet
  • 30+ expert sessions
  • 6 bootcamps and 13 courses
  • 13 hackathons
  • 20+ resourceful blogs

Community Highlights Overview:

  • Our major collaboration in 2023 involved LearnMachineLearning becoming a part of AI Planet, which expanded our community reach to over 350K+ AI enthusiats.
  • We hosted 30 expert live sessions in the Generative AI and Machine Learning space. These sessions provided our community the opportunity to learn directly from top industry speakers representing companies like Google, Intel, ZenML, LlamaIndex, and others.
  • Bootcamp & Courses: We hosted 6 bootcamps, including the most awaited LLM Bootcamp, and developed 13 courses, with a significant number in Spanish, thanks to the assistance of volunteers in Chile and Latin America.
  • Released Jupyter — AI Guru: To make the learning experience richer we released Jupyter AI, for the personalization of learning. Jupyter AI is integrated into our LMS.
  • Our ethos is focused on public learning, encouraging collaboration and self-improvement within our AI community. In line with this, we organized 13 hackathons, also referred to as AI Challenges, with the Global AI Hackfest and the IIT Guwahati challenges being especially prominent.
  • Additionally, we published 20 blog articles on AI-powered chatbots using Large Language Models and advanced Retrieval Augmented Generation (RAG) techniques, aimed at enhancing project-based learning.
  • AI Planet community partnered with InnoHub GenAI+. This immersive bootcamps was organized by Evenness and Women in AI, powered by Qualcomm, and supported by Tech communities, such as Mission Impact Academy, Women In DeepTech, Women in Tech Netherlands, and AI Planet.

Events are the gathering points for developers with similar mindsets, offering a vibrant arena for knowledge exchange and the cultivation of creative connections. Here are a few keyhights this year:

  • Hacktoberfest
  • Google Developers Group — Nepal
  • Ramaiah Institute of Technology — Bangalore
  • Digital Talent Summit & AI Con — Zurich

In celebrating the Open Source tradition, we participated in Hacktoberfest with our GenAI Stack. Notably, Hacktoberfest marked its grand 10th anniversary in 2023. Throughout Hacktoberfest, we hosted various sessions on getting started with GenAI Stack and Open Source, and on developing use cases for these technologies.

Our team actively engaged in multiple events, with one highlight being the DevFest hosted by Google Developer Groups. Tarun Jain, DevRel at AI Planet, was a featured speaker in Kathmandu, Nepal, where he presented on “Retrieval Augmented Generation and Fine Tuning.” He demonstrated an RAG application using the GenAI Stack. Additionally, he was honoured with an invitation from Ramaiah Institute of Technology, Bangalore, to inaugurate the Tensor AI Club as the Chief Guest.

AI Planet co-organized Digital Talent Summit, AI Con in collaboration with Global AI Hub, and C-level. We witnessed hundreds of top decision makers of leading companies and 13,000+ tech experts and talents from around the world. This was all possible thanks to collaboration and support from C-level, Global AI Hub, and guest speakers from AI Planet: Gaurav Anand and Srikanth Appana.

While being a remote first team, we always look forward to that time of the year where we all could come together and work under the same roof. This tradition is nothing new to us as we’ve been doing this the third successive year. Over the time, we’ve built solid bonding with our team members and set foundations for future during the team onsites. Some hightlights this year include:

  • Team Onsite in Hyderabad
  • Team visiting Visually Challenged Foundations
  • Our founder visited our extend team and partners at Istanbul

8 weeks of team onsite @ Hyderabad

This year, we organized our third team onsite in Hyderabad, India, bringing together our remote team members. The onsite event was not only enjoyable but also productive, as it facilitated decision-making for building reliable and scalable products. It was during this time that we released the alpha version of our GenAI Stack.

Volunteered for Proxmaq and visited visually challenged foundations

While at the onsite, AI Planet extended support to ProxMaq, a foundation for the focused on building inclusive society and supporting physically challenged with software and hardware products to overcome social and physical barriers. Our team visited the couple of visually challenged local foundations in Hyderabad along with Proxmaq founder and showcased a prototype of the smart lens powered by computer vision models. This experience in our team members terms is quite redefining and we look forward to continuing / accelerating our support to socially impactful organizations to bring more AI4Good solutions.

Team Onsite with our extended team and partners: C-Level, Global AI Hub & AI Business School

Moreover, our founder had the incredible opportunity to meet with the leadership teams of AI Business School, Global AI Hub, and C-Level in Istanbul. Engaging with these leaders was inspiring, providing valuable insights into their operational strategies and offering a chance to learn from their experiences. Many thanks to our investors, advisor and staunch supporter Selçuk Boydak for the invitation and hosting us.

Finally, listing some of the key collaborations, achivements and progress we’ve made this year:

  • fit4start graduation
  • Selection for Google for Startups, Microsoft for Startups
  • Specifically for GPU Compute: NVIDIA inception program and Luxprovide Initiate program
  • Our presence at WebSummit, ICT Spring, Vivatech, and Slush, where we’ve been featured and delivered talks among larger audience.

A major milestone for AI Planet was graduating from the fit4Start program by Luxinnovation GIE. This journey provided us with an intense coaching phase, under the guidance of some of the industry’s most experienced entrepreneurs. The program also awarded us a grant of 150,000 Euros.

Additionally, AI Planet is a part of Google for Startups, Microsoft for Startups, Nvidia Inception, and Luxprovide Initiate program. These affiliations have been crucial in developing our AI Marketplace and providing computational support for deploying state-of-the-art models.

Last but not least, we have made appearances at notable events such as Vivatech, WebSummit, Slush, and AI Con by C-level Switzerland partners. We also had the opportunity to showcase our product at booths in WebSummit and ICT Spring Luxembourg.

As we embrace the new year, we anticipate not only growing our existing partnerships but also exploring new opportunities, all while continuing to democratize AI (education, development and accessibility). The journey ahead promises to be an exciting blend of technological advancement and community growth. Thank you for being part of the AI Planet community and journey. Wish you a very happy new year!

Source link

Latest stories