Innovative thinkers across Africa came together for the 2024 Intel® Liftoff Hackathon for African AI Startups. It was five days filled with creative thinking and pushing the boundaries on innovation in AI.
Attracting teams from various backgrounds, the diversity fostered a rich exchange of ideas and solutions, highlighting the global nature of AI innovation.
Solutions addressed real-world challenges such as healthcare diagnostics, agricultural optimization, financial technology, smart city development, and education technology, showcasing the versatility and transformative potential of AI.
The winning submissions demonstrated the capabilities of generative AI and LLMs in creative and innovative ways. The overall impact of the event on the African AI community was significant, and here we highlight the overall winners of the challenges. Let’s dive in!
Submission 1 - FlexiSAF
Track 1:
LLM Model: Intel/neural-chat-7b-v3-1
Embedding Model: sentence-transformers/all-MiniLM-L6-v2
Vector Database: LanceDB
Document Extraction and processing: LlamaIndex
Dynamic context management and accurate relevant responses: Retrieval-Augmented
Generation (RAG)
We have built a chatbot that guides students, through selecting curriculums, pdf files and lecture books, provides clear explanations, answers questions, and creates interactive quizzes. been supportive, engaging, and motivational. It starts each interaction by greeting the student and asking how it can assist with their studies. staying within the context of the curriculum and uploaded files, if user ask about something that is out of context, it guides them back the topic or asked them to add the curriculum for that content.
Overview:
This documentation details the development process, challenges, insights, and practical implementations of a chatbot assistant designed to help students learn by selecting curriculum and uploading content. The chatbot leverages the Retrieval-Augmented Generation (RAG) model for dynamic context management and accurate, relevant responses. The database solution employed is LanceDB, with LlamaIndex used for efficient data retrieval. The application is designed to scale for massive user bases and adheres to production standards, aligning with Flexisaf's product distinction.
Development Process
Environment: Implemented within IDC simple inference LLM notebook, leveraging the provided Gen AI resources.
Database: Choose LanceDB for its efficient data retrieval and update capabilities. We have also experimented with custom vector store
Data Processing library: LlamaIndex, we experimented with both langchain and LlamaIndex, they both show promising results in data extraction and chunking.
Data Ingestion: Uploaded curriculum content and structured it using LlamaIndex to facilitate efficient retrieval.
Context Management: Used the RAG implementation to dynamically update and maintain relevant context based on user interactions and curriculum content.
Response Accuracy: Evaluated the chatbot's responses for accuracy and relevance using test scenarios, chat sessions will be seen below.
Scalability: This implementation is designed to scale for massive user bases and adheres to production standards, aligning with Flexisaf's product distinction.
The team at Flexisaf have benefited from the seamless integration and user-friendly nature of the Intel Developer Cloud and believe it has significantly enhanced productivity and creativity.
“Our experience at the Intel® Liftoff Hackathon was nothing short of transformative. Leveraging the power of the Intel® Developer Cloud with its cutting-edge GPU resources allowed us to explore and experiment with different AI models seamlessly. The platform's ease of use and robust capabilities were truly impressive, enabling us to push the boundaries of our project. Winning the hackathon was a testament to the support and technology provided by Intel, and it has inspired us to further innovate in the AI space.”, says FlexiSAF CEO Faiz Bashir.
Submission 2 - MicroFuse Technologies
Story Title: The Mother Tree
Story Summary: It encapsulates the journey of Lara and the enduring impact of her actions, symbolized by the tree that thrives and supports life long after her initial gesture of planting it.
A young girl, Lara plants a seedling in a barren field and years later, it turns to a flourishing oasis for a thriving community.
This story conveys a message of hope and the impact one person can have on the environment, inspiring viewers to believe in the power of small actions leading to significant change.
Development Process:
- I modified the given code and added delay by calculate the number of frames to repeat based on the video's frame rate and the desired delay duration and write each image multiple times to the video file.
- I added a transition effect by calculating the number of frames for the transition based on the frame rate and the desired transition duration and generating the frames for the transition by blending consecutive images.
Frame rate: Set to 10 frames per second (FPS) for smoother transitions.
Image display duration: Set to 1 second.
Transition duration: Set to 1 second.
Blend_images function: This function blends two images with a given alpha value, creating a smooth transition effect.
Main loop:
- Each image is written to the video file for the specified duration (num_frames_per_image).
- If it's not the last image, a transition is created by blending the current image with the next image over the specified number of frames (num_frames_for_transition)
Challenges:
The image_to_image notebook was not accepting a lot of external urls.
Insights:
It was quite interesting generating this video clip by just codes and not using any external software, it gave me the flexibility to tweak the process anyhow I wanted.
The MicroFuse team believes the Intel® Liftoff Hackathon was not just about the competition; it was about a remarkable journey of learning and growth. They believe the supportive community and collaborative atmosphere made it a unique experience. “The Intel Liftoff Hackathon for African AI Startups was a transformative experience, offering unparalleled innovation and collaboration opportunities. With mentorship from Intel experts and access to cutting-edge Intel technologies, our team accelerated project development and connected with potential partners. Winning the hackathon validated our hard work and creativity, but the most rewarding aspect was the knowledge and experience gained. The event was a catalyst for growth and the future of AI in Africa. We are grateful to Intel and the organizers for hosting this pivotal event”, says CEO and Founder, Emmanuel Ugwu.
Submission 3 - ToumAI
Motivations
"In Africa, when an old man dies, a library burns," said Senegalese writer Amadou Hampâté-Bâ. This quote perfectly illustrates the importance of elders in preserving the cultural heritage of the continent and passing down knowledge. Given the fleeting nature of life, it is crucial for each of us to consider ways to preserve these heritage elements that are part of our DNA.
Today, with the development of AI and the emergence of LLMs, it is possible to contribute to this endeavor. The goal of our project is precisely in this direction. We aim to leverage the power of such technologies to preserve African culture through tales.
More specifically, our solution, "GrandMaBOT" is based on a collection of tales extracted from various sources and different African countries, reflecting each country's folklore. The goal is to draw inspiration from these tales to generate custom-made original stories that respect African traditions and customs.
Given the scarcity of solutions that include African languages and the need to democratize AI in our continent, we wanted to make this solution multilingual to facilitate its use by people who do not necessarily understand foreign languages. To this end, prompts can be written in Wolof, Moroccan Arabic, Swahili, Bambara, etc., and receive text in these languages in return.
Technical Details
Our architecture is structured as follows:
Scraping and monitoring tales from multiple sources (PDF, HTML, raw data, etc.) and various countries (Mali, Morocco, Algeria, etc.).
Processing and structuring this data (grouping by tale, country, language, etc.).
Translating non-English data into English.
Implementing a RAG architecture using Falcon.
Fire! (Conversation with GrandMaBOT).
For the ToumAI team, even though there were challenges and new tech to adapt to, it was still a worthwhile experience.
COO Yousef Rahmani says, “This hackathon was a rich experience for our team both on the human side as well as on the tech side. We had fun and learnt a lot especially about the Intel platform, LLMs etc. which is giving us new perspectives for our startup”.
Submission 4 - UTU Technologies
Setup
Model open_llama_3b_v2, only changed the Interaction param to “Interact with context”.
Changed the sample_prompt in warmup_model to
“”"A dialog, where User interacts with a Bot.
User and Bot both act as characters in a sci-fi story. Bot’s character is called Amani.
User: Hi Amani, what are you doing here? I’ve never expecter to meet you here!
Bot: Oh, I’ve moved to Titan last year.
User: Wow, all worlds are small. Come let’s go have a drink. You have to tell me all about this place.“”"
Changed the prepended text for user_input in prepare_input to
“Below is the next bit of dialogue from User’s character. ”
“Write a response that appropriately provides Bot’s character’s next response.\n\n”
CEO, Jason Eisen had the following to say, “As an African AI company that cares mostly about improving trust on the internet globally, we could not be more excited to collaborate with Intel, one of the most trusted brands on the planet, to bring better trust to the next internet.”
Submission 5 - Alcare
Title: The Lost Artifact
Image1: "A young explorer named Aria discovers an ancient map hidden in a tree hollow in a mysterious forest, surrounded by dense trees and mist. The scene is filled with an adventurous and magical atmosphere."
Image2: "Aria, the young explorer, navigates through a perilous mountain pass, facing treacherous cliffs and wild animals. The environment is harsh and the atmosphere is tense as she follows the ancient map."
Image3: "In a hidden jungle temple, Aria encounters an ancient guardian spirit as she approaches an artifact. The scene is magical, with the spirit testing her worthiness amidst glowing runes and mystical light."
Image4: "Inside the temple chamber, Aria holds the ancient artifact, which emits a bright, enlightening light. The surroundings are filled with ancient carvings and symbols, creating a triumphant and awe-inspiring atmosphere."
Image5: "Aria returns to her village and shares the wisdom of the artifact with her people. The villagers celebrate in a joyful and hopeful atmosphere, the ancient knowledge bringing prosperity and happiness."
Conclusion
The Intel® Liftoff Hackathon for AI startups was a remarkable success, highlighting the innovative potential of diverse teams. It also showcased how AI is transforming all sectors, not only on a global scale, but specifically on the African continent.
Now that the Hackathon is over, winners will officially be onboarded to the Intel® Liftoff program, where they will continue to receive the necessary support and guidance. This includes access to mentors who will help them refine their solutions and scale their startups. The program offers resources, networking opportunities, and potential funding to accelerate growth and impact.
Be sure to follow Intel® Liftoff for future updates and opportunities in AI.
As a Software Tools Ecosystem Specialist at Intel, I’ve had the privilege of working on the dynamic GenAI initiative. My focus is on driving engagement with the software developer audience. I'm a proud team member of the Intel® Liftoff for Startups and the Developer Engagement, Relations, and Studio team.