A Tiny Model is a specialized paradigm in the realm of machine learning, characterized by its compactness and efficiency.Unlike Large Language Models (LLMs) that require substantial computational resources, and vast storage, and can be expensive to run on private clouds, Tiny Models are designed to be lightweight and nimble. Their architecture is such that they can be hosted, trained, fine-tuned, and employed on consumer-grade computers, demanding minimal resources.The quintessential appeal of Tiny Models lies in their ability to deliver very good results for specific domains or tasks, without the overheads typically associated with their larger counterparts. They present a viable and cost-effective alternative for organizations seeking domain-specific solutions within their premises, without compromising on accuracy or performance.Why Choose Our Tiny Transformers?1. Specialization in Sub-1 Billion Parameter Models Efficiency and Accessibility: We focus on developing very small models, directly addressing the computational and financial constraints of small businesses. 2. Custom Pre-Training with Business-Specific Data Personalization and Competitive Differentiation: We offer custom pre-training with each business's own data. Our highly tailored datasets for training to ensure that the models are not just general-purpose, but are specifically optimized for the nuanced needs of each business.3. Multi-Class Classifiers and Autoregressive Generation Versatility in Applications: By combining trees of multi-class classifiers with autoregressive generation for Span QA applications, we provide a versatile solution that can be adapted to a wide range of use cases, from customer service automation to personalized content creation.4. Consumer-Grade Hardware Compatibility Cost-Effectiveness and Scalability: We ensure our models can run on consumer-grade hardware with small GPUs, lowering the barrier to entry for small businesses to adopt AI technologies.

At MLBootcamp.AI, we strive to bridge the gap between theoretical knowledge and practical application in the machine learning domain. Our platform offers an end-to-end learning experience, ensuring every individual can comprehend the intricacies of ML projects and apply them in real-world scenarios.We envision a future where every enthusiast, regardless of their background or prior experience, possesses the hands-on skills to harness the power of machine learning for real-world applications. We believe in a holistic learning approach that transcends traditional boundaries, emphasizing practical experience as much as theoretical knowledge. As the realm of machine learning continues to evolve, we aim to remain at the forefront, ensuring our community is always equipped with the latest tools and methodologies to drive impactful solutions across various industries.MLBootcamp.AI was born from a vision to democratize machine learning education. In a landscape dominated by disjointed tutorials and isolated lessons, we recognized the urgent need for a structured and unified learning experience. Our emphasis isn't just on isolated modeling techniques; we are passionate about introducing learners to the comprehensive world of end-to-end machine learning projects.While we might be at the beginning of our journey, our aspirations are clear. We aim to be a leading platform, bridging the gap between theoretical knowledge and practical implementation. Our courses are meticulously designed to ensure every student doesn't just learn but also applies the knowledge, building a robust portfolio in the process. This portfolio-centric approach is our way of ensuring that our learners are not only well-versed in machine learning concepts but also ready to demonstrate their real-world application prowess.

VisiTrack Corp, a No-Code Field Service platform. VisiTrack is specialized in Digital Transformation, from the conversion of paper documents to electronic documents, to the automation of workflows, Internet of Things, Artificial Intelligence, Big Data and Machine Learning, tailored to the vision and needs of your company.

ServiTrack is a Field Service App for the service technicians. Developed from the ground up with the assets in the center, and the interactions with Jobs, Technicians and Customers. It is AI-Supported and IoT-Enabled Field Service for the service industry. Can be integrated with other back-end systems.

AI RISK ASSESSMENT​To help you identify potential issues and optimize your AI system's performance.MODEL ACCURACY ASSESSMENTEnsure the accuracy of your AI models through our comprehensive assessment with actionable insights for improvement​.EFFICIENCY OPTIMIZATIONTo help you streamline your AI system, reducing operational costs and improving overall efficiency.

As a member of the AI/LLM Advisory Board at Permio AI, I am involved in steering the integration and ethical deployment of artificial intelligence to revolutionize the construction permitting process. Permio AI focuses on enhancing efficiency and reducing submittal errors in construction projects through advanced AI solutions. In my role, I contribute to the strategic direction of leveraging AI alongside human expertise to optimize permit workflows, thus significantly reducing time to construction start and improving overall project execution.Our AI smart assistant, Mio, exemplifies our commitment to innovation by automating complex tasks such as zoning research, form completion, error detection, and rapid response to queries, freeing up essential resources and ensuring projects proceed smoothly and swiftly.

Model: ChatBOT with Canada's immigration information, SentenceTransformer trained to produce answers based on ad-hoc questions.Project title: Empowering Canada's Immigration Applicants with Accurate AI Chatbot Assistance.Every year, millions of people across the world apply to come to Canada as an immigrant. Very few make it to the final list. Some fail due to genuine eligibility criteria while others fall prey to misinformation and scams. This happens mainly because people do not know how to go about using the CIC immigration website by the govt of Canada. A lot of pseudo-agents give false hopes to desperate people and scam them of money.This project will help solve the ambiguity related to the entire immigration process, be it related to the documents needed, the funds required, any change in the status of an applicant, etc. The applicants will no longer need to rely only on the agent’s information but can rather use this chatbot and get their queries resolved.

LLM Ops - Large Language Models in ProductionObjective: Design, assemble, and operate production LLM apps with industry-standard tools.- Understand LLMs and their implications for product development in your organization or startup- Build complex LLM apps with the most popular LLM Ops frameworks including LangChain and LlamaIndex- Enhance your portfolio by contributing Python code to an open-source production LLM application- Build, ship, share, benchmark, and improve your own end-to-end production-grade LLM application- Work with other talented and driven builders to bring your next passion project or Generative AI startup idea to life

Reinforcement Learning with Human Feedback on Text Summarization, step-by-step, from concepts to code!Covering each of the core components of RLHF, including:- Base model- Rewards model- Proximal Policy Optimization model - EvaluationBreaking down all key aspects of #RLHF:- ❓ the What, How, When and Why- 🤖 end-to-end workflow- 🏋‍♀️ training the #RewardModel on #HumanFeedback (hands-on)- 🔧 fine-tuning using #ProximalPolicyOptimization (hands-on)- 🔥 tech stack: #TRL library, #BeRT, #T5_base, #Peft & #LoRA, #HuggingFace

As a mentor for the DeepLearning.ai courses in machine learning and deep learning, I play a critical role in helping students succeed in their learning journey. The main responsibilities involve guiding, supporting, and motivating students to ensure they effectively grasp the course content, and develop their skills. Here are some key aspects of the mentor's role:Subject Matter Expertise: Mentors possess a deep understanding of the course material, related tools, frameworks, and techniques in machine learning and deep learning. They are able to provide accurate, clear, and concise explanations of complex concepts to students.Student Support: Mentors are available to help students when they face difficulties or have questions about the course content. They provide personalized assistance to address specific learning needs, troubleshoot issues, and ensure students stay on track with their learning goals.Assignment and Project Review: Mentors provide constructive feedback on assignments and projects, helping students identify areas for improvement and suggesting ways to enhance their work. This feedback is crucial for students to refine their skills and gain confidence in their abilities.Encouragement and Motivation: Mentors inspire and motivate students to stay committed to their learning journey, overcome challenges, and achieve their goals. They celebrate student successes and help build a positive learning mindset.Continuous Improvement: Mentors collaborate with the DeepLearning.ai team, providing feedback on course content, assignments, and projects to ensure the curriculum remains relevant, up-to-date, and effective for students.

Model: Computer vision - CNN multi-class classifier train from scratch to identify one of 6 classes (coffee leaf disease), StyleGAN trained from scratch, Transformer trained from scratch for multi-class classification. Project: Classification of Plant Diseases in Brazilian Agriculture Using Computer Vision and Machine LearningBrazil, known for its vast agricultural landscapes, faces significant challenges in monitoring and diagnosing plant diseases that can negatively impact crop yields.One of the key objectives is to develop accurate and efficient disease detection models that can identify specific symptoms and patterns associated with various plant diseases. By training these models on diverse datasets encompassing different crops and disease types, researchers aim to create robust and adaptable systems that can accurately classify and quantify diseases.However, several challenges persist in this field. The high variability of environmental conditions, such as lighting, weather, and plant growth stages, poses a significant hurdle for accurate disease detection. Furthermore, the diversity of Brazilian crops and diseases requires extensive research and development efforts to ensure generalization and applicability across different regions and conditions.A machine learning solution using computer vision can significantly contribute to solving the problem of plant disease management in Brazilian agriculture and have a positive impact on the local community. This collaborative effort can help to the creation of robust disease detection models that are tailored to the specific needs of Brazilian crops and farming practices. By accurately and efficiently identifying diseases, these models will empower local farmers with timely information for proactive decision-making, allowing them to implement targeted treatments, reduce crop losses, and optimize resource usage.

Model: DizirBERT transformer fine-tuned for a binary-classification task to detect hateful language.Classification of social media content in Algerian Dialect using NLP and Machine learning.Focus: Detecting Hate Speech on Social Media in the Algerian Dialect with NLPIn the Algerian context, the development of NLP-based solutions is hampered by the general use of dialect in conversation on social media. Thus, there is a need to develop customized NLP solutions that can handle dialect-based text content.

Models: Recommender systems: Content-based recommender system and Collaborative-based recommender system. Output: personalized recommendations for users based on 4 pillars: Nutrition, Activity, Rest, and Beauty.Our goal is to create a comprehensive wellness measurement and recommendation system that can help individuals improve their physical, mental, and emotional health across four key areas: physical activity, stress management, nutrition, and self-perception or self-esteem.- Develop a system that can measure an individual’s wellness across four key pillars: physical activity, stress management, nutrition, and self-perception or self-esteem.- Create a recommendation system that provides personalized wellness content based on the user’s individual needs and context, including their calendar, location, and social network.- Prompt users at appropriate times throughout the day to engage in wellness activities, such as simple exercises or activities that can help re-energize and refocus them.- Increase user engagement with wellness content by providing recommendations that are relevant and tailored to their needs and circumstances.- Measure the effectiveness of the system in improving users’ well-being over time and make necessary adjustments to improve its performance.- Make the system flexible enough so it can be updated based on user feedback and evolving trends in the wellness industry.

Model: ChatBOT - predictive model, recommender model, RAG to generate coherent recommendations.PROJECT FINISHED SUCCESFULLY: The goal of this project is to develop a chatbot application that helps the citizens in the Ile-De-France by providing them with reliable and accurate information about alternative transportation on strike days. During a strike day, users may struggle to find reliable and accurate information about alternative transportation in Ile-De-France. The deliverable will be a conversational AI Chatbot for alternative transportation during strikes in Paris, France. Bike, train, buses, taxis, walking, and any other transportation alternative will be included.

Model: Recommender system - both Content-based and Collaborative-based recommender systems. Output: Personalized recommendations of where to buy and what groceries to buy based on user's preferences.PROJECT FINISHED SUCCESFULLY: Developing a Recommended System for Grocery Shopping in Berlin.Objective: Develop a system based on AI to help customers make informed decisions about where to shop for their groceries.With so many options available, it can be overwhelming to know which store has the best deals, selection, and quality for the customer’s specific needs and preferences.The recommended system can help to alleviate this problem by providing personalised recommendations based on factors such as location, product availability, price, and customer reviews.By doing so, customers can save time and money while also ensuring they are satisfied with their grocery shopping experience.

Model: Prediction of churn based on customer behavior. Regression model developed as part of the solution. Several models used: trees, DNN, Logistic regression, etc. Details below.PROJECT FINISHED SUCCESFULLY: E-Commerce Customer Churn Prediction: Build ML Model for Customer Churn Prediction.Challenges:Unbalanced dataset containing only 17% of churn cases and a small sample size of 5600 records. Models:Decision Trees, Deep Neural Networks (DNN), Random Forest, Logistic Regression, K-Nearest Neighbors (KNN), and Light Gradient Boosting Machine (LGBM). Achievements:The LGBM model achieved an accuracy of 0.99, recall of 0.98, precision of 0.93, F1 score of 0.96, and AUC of 0.98. Next steps:1. Provide the source code to the ecommerce sites for them to integrate it, or 2. Deploy a web service that would allow ecommerce sites to submit their data and get results, making sure that all the data is properly anonymized and secure and private to each site.

Model: Several models implemented looking for the best solution: LSTM, DNN, DNN+LSTM with ragged tensors, CNN, CNN_LSTM, Transformer. Details and results below.PROJECT FINISHED SUCCESFULLY: Google - Isolated Sign Language Recognition: Enhance PopSign's educational games for learning ASLGoal:The goal is to classify isolated American Sign Language (ASL) signs by creating a TensorFlow Lite model trained on 90k+ parquet files of labeled landmark data extracted using the MediaPipe Holistic Solution.This work may improve the ability of PopSign* to help relatives of deaf children learn basic signs and communicate better with their loved ones.Threshold score: 60Final score: 62Top leaderboard: 79Teams: 978Models implemented: LSTM, DNN, DNN+LSTM with ragged tensors, CNN, CNN_LSTM, Transformer.

Yaqui es la plataforma digital que permite conectar a proveedores con compradores en un solo lugar, brindándoles soluciones de Marketplace como:- Catálogos virtuales- Carrito de compras- Opciones de Pago (Pasarela de Pagos)Adicionalmente ofrece:- Herramientas de colaboración y comunicación analítica de eficiencia de mercadeo digital (reportes y estadísticas de visitas a cada comercio virtual).- Educación y asesoría a clientes en su paso a la transformación digitalYaQui ofrece:- Ecommerce de productos- Ecommerce de servicios- Rueda de Negocios- Plataforma de reservas- Pasarela de pago- Centro de Aprendizaje- Plataforma Logística y Domicilios

A STRATEGIC APPROACH TOWARDS A GREAT EDUCATION - From kindergarten to 12th grade IvyNue helps parents and children to achieve outstanding results through their school years.

STEAM Gym was a robotics, science, and art academy for children from 5 to 13 years. The academy was located Cali, Colombia.STEAM Gym is a place where kids will develop skills, critical thinking, and healthy competitive spirit, while learning robotics, coding, science, and expressing themselves through art.1. Kids will build and program robots. Each project is designed to solve a problem with a robot.2. Kids will do science experiments.3. Kids will do art projectsSTEAM Gym follows a methodology based on a sequence of projects, where each project involves new concepts and new challenges with increasing complexities. Each child follows its own path while working in a collaborative environment.STEAM: Science, Technology, Engineering, Art and Math, is a new global trend in education. Its goal is to introduce kids to science from different angles. STEAM Gym is a chain of academies where kids will learn about robotics, programming, and science, and will also have ART classes to balance the science with liberal thinking. STEAM Gym’ main focus is in Robotics and Programming. Science and Arts become complementary and are used to break the routines of the robotics and programming tracks. Our methodology also includes sessions when kids present their projects to their peers and parents. Some sessions include competition among the participants.