META GPT

In this article , we present MetaGPT, a pioneering multi-agent framework incorporating real-world expertise based on SOPs. Firstly, each agent is identified by a descriptive job title, allowing the system to initialize with an appropriate role-specific prompt prefix. This embeds domain knowledge within agent definitions, rather than simplistic role-playing prompts. Secondly, we analyze efficient human workflows to extract SOPs encapsulating procedural knowledge required for collaborative tasks. These SOPs are encoded into the agent architecture through role-based action specifications.
Thirdly, agents produce standardized action outputs to enable knowledge sharing. By formalizing artifacts that human experts exchange, MetaGPT streamlines coordination between interdependent roles. Finally, a shared environment connects agents, providing visibility into actions and shared access to tools and resources. This environment contains all messages exchanged between agents. Additionally, we provide a global memory pool to store all collaboration records where each agent can subscribe to or search for the information they require. Agents can extract past messages from
this memory pool for additional context. This design enables agents to actively observe and pull pertinent information, which is a more efficient approach compared to passively receiving data via dialog. The environment mirrors human workplace infrastructures that facilitate team collaboration.

The MetaGPT framework facilitates the creation of various specialized role classes, such as ProductManager, Architect, and others, which inherit from ?the base Role class. A base role class is characterized by a set of key attributes: name, profile, goal, constraints, and description. Specifically, Profile represents the domain expertise of the role or job title.

In the MetaGPT framework, we define key components like Environment, Memory, Roles, Actions and Tools in detail,and develop foundational capabilities related to collaboration.

• Environment – Offers a collaborative workspace and communication platform for agents.

• Memory – Facilitates agents in storing and retrieving historical messages and context.

• Roles – Encapsulate specialized skills, behaviors, and workflows based on domain expertise.

• Actions – Procedures executed by agents to accomplish subtasks and generate outputs.

• Tools – Collective utilities and services that agents can utilize to enhance their capabilities

For instance, an Architect’s profile might encompass software design, while a ProductManager’s profile could concentrate on product development and management. Goal signifies the primary responsibility or objective that the role seeks to accomplish. A ProductManager’s goal might be expressed in natural language as efficiently creating a successful product. Constraints denote limitations or principles the role must adhere to when performing actions. For example, an Engineer could have constraints to write standardized, modular, and maintainable code. The constraints might be articulated as The code you write should conform to code standards like PEP8, be modular, easy to read, and maintain. Description provides additional concrete identity to help establish the role more comprehensively. Role initialization in the MetaGPT framework employs natural language to thoroughly describe the responsibilities and constraints of each role. This not only aids human understanding but also directs the LLMs to generate actions that align with the role’s profile, thereby rendering each agent proficient in its role. We define this process as anchor agents, which assists humans in encoding domain-specific responsibilities and capabilities to an LLM-based agent while also adding behavior guidance on expected functions.

The comprehensive role definitions provided by the MetaGPT framework enable the creation of highly specialized LLM-based agents, each tailored for specific domains and objectives. This not only introduces a layer of behavior guidance based on expected functions but also facilitates the creation of diverse and specialized agents, each expert in its domain. This leads to the development of more effective and efficient LLM-based agents capable of handling a wide

range of tasks. In MetaGPT, intelligent agents not only receive and respond to information, but they also observe the environment to extract critical details. These observations guide their thinking and subsequent actions. Finally, significant information extracted from the environment is stored in memory for future reference, effectively making every agent an active learner within the system. They take on specialized roles and follow certain key behaviors and workflows:

Think & Reflect Roles can retrieve role description (position) and prefix to frame thinking, and then reflect on what needs to be done and decide next actions, via _think() function. “Think first, then act” – carefully deliberate before replying

Observe Roles can observe the environment and think/act based on observations using the _observe() function. They watch for important information and incorporate into memory to enrich their contextual understanding and informing future decisions. Broadcast Messages Roles can broadcast messages into the environment using the _publish_message() function. These messages contain details about current execution results and related action records, for publishing and sharing information. Knowledge precipitation & Act Roles are not only broadcasters but also recipients of information from their

environment. Roles can assess the incoming messages for relevancy and timeliness, extract relevant knowledge from shared environment and maintain an internal knowledge repository to inform decisions.They execute actions via consulting to LLM with enriched contextual information and self knowledge. Execution results are encapsulated as Message while norm artifacts are shared by the environment. State Management Roles can track their actions by updating their working status and monitoring a to-do list. This enables a role to process multiple actions sequentially without interruption. When executing each action, the role first updates its status to busy. After completing the action, it marks the status as idle again. This prevents other actions from

interrupting the flow. This is a crucial capability in role design, making roles more human-like. It grants roles more natural execution dynamics grounded in real-world human collaboration phenomena.

In summary, the MetaGPT framework offers a versatile and powerful approach to designing and implementing intelligent agents with specialized capabilities. These agents can effectively collaborate, learn, adapt, and perform various tasks, making them valuable assets in a wide range of applications and domains.

The comprehensive role definitions provided by the MetaGPT framework enable the creation of highly specialized LLM-based agents, each tailored for specific domains and objectives. This not only introduces a layer of behavior guidance based on expected functions but also facilitates the creation of diverse and specialized agents, each expert in its domain. This leads to the development of more effective and efficient LLM-based agents capable of handling a wide

range of tasks.In MetaGPT, intelligent agents not only receive and respond to information, but they also observe the environment to extract critical details. These observations guide their thinking and subsequent actions. Finally, significant information extracted from the environment is stored in memory for future reference, effectively making every agent an active learner within the system. They take on specialized roles and follow certain key behaviors and workflows:

Think & Reflect Roles can retrieve role description (position) and prefix to frame thinking, and then reflect on what needs to be done and decide next actions, via _think() function. “Think first, then act” – carefully deliberate before replying .Observe Roles can observe the environment and think/act based on observations using the _observe() function. They watch for important information and incorporate into memory to enrich their contextual understanding and informing future decisions. Broadcast Messages Roles can broadcast messages into the environment using the _publish_message() function. These messages contain details about current execution results and related action records, for publishing and sharing information. Knowledge precipitation & Act Roles are not only broadcasters but also recipients of information from their environment. Roles can assess the incoming messages for relevancy and timeliness, extract relevant knowledge from shared environment and maintain an internal knowledge repository to inform decisions.They execute actions via

consulting to LLM with enriched contextual information and self knowledge. Execution results are encapsulated as Message while norm artifacts are shared by the environment. State Management Roles can track their actions by updating their working status and monitoring a to-do list. This enables a role to process multiple actions sequentially without interruption. When executing each action, the role first updates its status to busy. After completing the action, it marks the status as idle again. This prevents other actions from interrupting the flow. This is a crucial capability in role design, making roles more human-like. It grants roles more natural execution dynamics grounded in real-world human collaboration phenomena.

In summary, the MetaGPT framework offers a versatile and powerful approach to designing and implementing intelligent agents with specialized capabilities. These agents can effectively collaborate, learn, adapt, and perform various tasks, making them valuable assets in a wide range of applications and domains.

USE CASE

The Use Case is a standard tool in the BA toolkit, and like most tools they come in different shapes and sizes for different jobs.

During Business Analysis Planning it is important to identify the appropriate use case format for the project needs. At the high end we have a multi-page highly regulated requirements template. The base model is a single actor and basic flow. Between these is the textbook version with complex flows. This article looks at these use case formats and design influences to consider during your Business Analysis Planning.

imple Flow use cases describe the interaction of a single actor along one basic flow. This format can be used as the starting point for a more complex effort, as a placeholder for planning purposes, or as the final deliverable for a simple project. This format is similar to a user story, with post conditions instead of benefits.

Complex Flow use cases include alternate flows, triggers, business rules, post and exit conditions. The degree of complexity should be determined by the stakeholder needs and perspectives. For business users the documentation must be readable so that they can follow and agree that you have captured their needs correctly. For the architects and developers, the details must be specific enough to translate into units of development. For the testing teams the documentation must provide clear inputs and outputs of success so that they can formulate test cases.

Highly Prescribed use cases have a multi-page template with approved instructions to complete each section. There are projects which require such discipline, but I have also worked on projects where the template became the end rather than the means, and the degree of detail slowed down the project and produced wasteful documentation.

Use cases describe the functional requirements of a system from the end user’s perspective, creating a goal-focused sequence of events that is easy for users and developers to follow. A complete use case will include one main or basic flow and various alternate flows. The alternate flow — also known as an extending use case — describes normal variations to the basic flow as well as unusual situations

A use case should:

Organize functional requirements.
Model the goals of system/actor interactions.
Record paths — called scenarios– from trigger events to goals.
Describe one main flow of events and various alternate flows.
Be multilevel, so that one use case can use the functionality of another one.
How to write a use case
There are two different types of use cases: business use cases and system use cases.

Here are some significant AI use cases impacting major industries

1. Healthcare

The contribution of technology giants like Microsoft, Google, Apple and IBM in the healthcare sector holds significant importance for the healthcare industry. AI is currently employed in a wide range of healthcare services, including data mining for identifying patterns and carrying out highly accurate diagnoses and treatment of medical conditions. In addition, it is used in medical imaging, medication management, drug discovery and robotic surgery.

2. Retail and e-commerce

Retail and e-commerce stand out as the industries where the application of AI is most readily apparent and observable, making its impact tangible to the majority of end-users. Because of stiff competition in the market, retail organizations always look out for techniques to find patterns in consumer behavior so they can align their business strategy with consumer needs to outsmart competitors.

AI has played a pivotal role in empowering these organizations to effectively meet and exceed their customers’ evolving needs and expectations. The product recommendations on your Amazon account are nothing but an application of complex AI algorithms to determine which products you are more likely to buy

3. Food tech

AI has numerous applications in the food industry. Have you ever wondered what it would be like if a robot could brew a cup of tea for you? Well, Hi Arya, a food-tech company, in collaboration with LeewayHertz, has built a robotic tea maker based on AI and IoT capabilities. The smart tea maker enables users to create their own recipe from a web interface, mobile app, and the machine itself.

Essentially, users command the robot to prepare tea using the web interface, machine, or mobile app. As soon as a user places their order, the machine starts preparing the tea, and the user can watch their tea being prepared.

AI development has touched the industrial food processing sector as well. For instance, a firm named Tomra Systems ASA has developed AI-based food sorting equipment targeting the french fries, peeled potatoes, and the diced and ground meat markets. Tomra’s food processors help food-processing companies automate food analysis tasks such as measuring the size, shape, and color of french fries or analyzing the fat content in meat.

4. Banking and financial services

The banking and financial services industry is undergoing a massive transformation due to the onset of AI applications. AI uses cases in this space are plenty. In many scenarios, human agents are being replaced by intelligent software robots for processing loan applications in fractions of a second. Similarly, robo-financial advisors can sift through multiple levels of data in seconds to recommend the right investment decisions for customers.

These robo-advisors can also analyze your social media activities, emails and other personal data to identify the sectors and companies aligned with your needs and long-term goals.

Furthermore, AI-based chatbots are being deployed in the Insurance sector to improve the customer experience and create insurance plans and products based on customers’ data. AI-based software has also significantly reduced the claim processing time, thus helping insurance companies and customers

5. Logistics and transportation

The logistics and transportation industry has also benefitted significantly from AI-powered solutions. The use of machine learning has already transformed supply chain management, making it a seamless process. Many warehouses use AI-powered robots for sorting and packaging products in warehouses. Furthermore, AI algorithms are also increasingly used to find the quickest shipment route and support last-mile delivery.

In the transportation industry, self-driving vehicles will undoubtedly be the next big thing. Although they are still in the research and trial stage in many countries, AI-based self-driving will potentially replace manual driving and make driving on roads safer. Tesla, Uber, Volvo and Volkswagen are at the forefront of this innovation.

6. Travel

The travel industry derives significant benefits from the widespread use of AI-enabled chatbots. Chatbots are a proven means for improving customer service and engagement mainly because of their 24*7 presence and the ability to ensure instant resolution of queries.

Advanced AI algorithms are empowering chatbots with increased efficiencies, enabling them to provide highly accurate responses to customer queries. Many large travel organizations are turning to AI companies to build their own AI-based mobile apps and chatbots to improve the customer experience.

Furthermore, machine learning and predictive analytics help travel companies increase their conversion rates by identifying customer behavior and purchasing patterns.

7. Real estate

The application of AI in the real estate industry is opening up new opportunities for agents, brokers and clients alike. While agents are becoming more efficient and effective, brokers are becoming more strategic in their approach, and consumers feel more empowered than ever. AI-powered bots help brokers and agents efficiently cater to the needs of people looking to buy, rent or sell their properties.

AI can be employed in the real estate industry in the following ways.

Real estate professionals can use artificial intelligence to analyze market conditions, property prices, and other factors to determine property values, trends, and investment opportunities.
Using artificial intelligence, real estate documents, such as lease agreements, mortgages, and title deeds, can be managed and processed automatically.
The predictive modeling component of AI can be used to predict rental income, property prices, and other aspects affecting the real estate market.
Smart home technology, such as thermostats and security systems, can be integrated with AI to increase energy efficiency and security.
Also, AI-based chatbots can operate 24*7 and help real estate website visitors find answers to their queries, even during odd hours

8. Entertainment and gaming

From OTT platforms recommending personalized shows to viewers to video games enhancing their visuals and improving the gameplay experience for players, AI empowers entertainment businesses with intelligent capabilities.

In the film industry, AI is used to enhance digital effects in movies and perform diverse other tasks to save costs and speed up the pre and post-production process. For instance, Natural Language Processing (NLP) can be used to create movie scripts.

In the music industry, large companies like Apple and Spotify implement AI to understand users’ engagement patterns and recommend music to users based on their tastes. In music production, the AI-driven computer accompaniment technology enables a machine to compose real-time music in response to the performance of a live musician.

The gaming industry was one of the early adopters of AI, and its impact on the user experience has been profound. This means that within the gaming industry, AI is employed to govern the behaviors and actions of Non-player Characters (NPCs) who contribute to progressing the game’s narrative in a predetermined direction. AI-driven behavior modeling of such characters greatly enhances the gamer’s experience in the overall storyline.

9. Manufacturing

It is beyond doubt that the manufacturing industry is leading the way in the application and adoption of AI technology. In manufacturing, AI is employed across several lines and layers of operations, from workforce planning to product design, thus improving efficiency, product quality and employee safety.

In factories, machine learning and artificial neural networks are employed to support the predictive maintenance of critical industrial equipment, which can accurately predict asset malfunction. It helps the management take timely measures to restore the equipment and prevent costly unplanned downtime

10. Automotive

From self-driving cars to driver assistance systems and traffic prediction to improve safety and reduce traffic congestion, AI has several potential practical applications in the automotive industry. It can be used to power self-driving cars, helping them to make smart decisions, navigate roads and avoid obstacles. Driver assistance systems, such as adaptive cruise control, lane departure warning, and automatic emergency braking, can be enhanced with the help of artificial intelligence. Using cameras, radar, and other sensors, these systems detect and accommodate traffic conditions, contributing to safer and more efficient driving.

Using artificial intelligence, traffic signals, signs, and other infrastructure can be adjusted to respond to changes in traffic conditions to optimize traffic flow and reduce congestion on roads. By analyzing data from vehicles, AI can determine when maintenance will be needed, including when a component will fail. By scheduling maintenance before a problem occurs, automakers and fleet operators can reduce downtime and costs.

11. Media

AI has significantly impacted the media industry, offering various applications and use cases. It can enhance the media industry in the following ways:

Content recommendation: AI algorithms can analyze user preferences and behavior to provide personalized content recommendations, enhancing user experience and engagement.
Natural Language Processing: AI-powered language processing can be used to analyze and understand user-generated content, automate content moderation, and improve search functionality.
Video and image analysis: AI can be used for video and image analysis, such as object recognition, sentiment analysis, and facial recognition, enabling better content organization and targeting.
Fraud detection: AI algorithms can detect fraudulent activities, such as fake user accounts or click fraud, helping to maintain a safe and secure platform.
Predictive analytics: AI can analyze large volumes of data to predict user behavior, market trends, and audience preferences, assisting in strategic decision-making and content planning.

12. Education

AI in the education industry has transformed the way we learn, offering a multitude of use cases that have positively impacted the landscape. Some of these use cases are:

Personalized learning: AI can personalize educational content and adaptive learning paths based on individual student needs, improving engagement and learning outcomes.
Intelligent tutoring: AI-powered tutoring systems can provide personalized guidance, feedback, and assistance to students, enhancing their learning experience.
Automated grading: AI algorithms can automate the grading process for assignments, quizzes, and exams, saving time for educators and providing quick feedback to students.
Learning analytics: AI can analyze student performance data to discover patterns, trends, and areas of improvement, enabling data-driven interventions and personalized support.
Virtual assistants: AI-powered virtual assistants can answer student queries, provide educational resources, and assist in administrative tasks, enhancing the learning environment.

13. Fashion

AI has found compelling applications in the fashion industry, impacting various aspects of the value chain. One key area is trend analysis and forecasting. AI algorithms can identify emerging trends, consumer preferences, and fashion influencers by analyzing vast amounts of data from social media, fashion blogs, and online platforms.

AI-powered recommendation systems can analyze customer data, including purchase history, browsing behavior, and style preferences, to provide personalized product recommendations. This enhances the customer experience and increases sales and customer loyalty.

By leveraging machine learning algorithms, AI can accurately forecast demand by analyzing historical sales data, market trends, and external factors like weather patterns. This helps fashion brands, and retailers optimize inventory management, production planning, and logistics, reducing costs and minimizing wastage.

AI algorithms can analyze images and identify specific patterns, colors, or designs, enabling automated product categorization, tagging, and cataloging. This streamlines inventory management and improves searchability on e-commerce platforms. AI can also assist in detecting counterfeit products by comparing images and identifying discrepancies.

L’intelligence artificielle au cœur de la transformation durable des bâtiments

Responsables de 40 % des émissions carbone dans le monde, les bâtiments doivent accélérer leur transformation pour espérer s’approcher de l’objectif net zéro en 2050. Le chemin pour y parvenir est étroit, avec une chaîne de valeurs très fragmentée et un taux de rénovation trop faible. Néanmoins, les technologies de pilotage intelligent des bâtiments peuvent aider à opérer une transition positive à grande échelle.

L’impact climatique des bâtiments reste trop élevé

Selon la Global Alliance for Buildings & Construction des Nations Unies, le secteur du bâtiment et de la construction est un émetteur majeur, responsable d’environ 40 % des émissions mondiales de carbone. Près des trois quarts environ sont dus à la consommation d’énergie des bâtiments et un quart à leur construction, notamment en raison des matériaux utilisés. Alors que l’on s’attend à ce que la superficie des bâtiments dans le monde double avant 2050, l’empreinte des bâtiments menace d’épuiser le budget carbone restant. La chaîne de valeur fragmentée du secteur constitue par ailleurs un défi majeur, car il est plus difficile d’opérer des changements positifs à grande échelle et rapidement.

Selon un rapport du BPIE, l’impact climatique des bâtiments reste ainsi trop élevé, et les nouvelles normes visant à réduire les émissions carbone dans le secteur ne suffiront pas à atteindre les objectifs climatiques de l’UE, en l’occurrence une réduction de moitié des émissions directes d’ici 2030 et un parc immobilier net sans carbone d’ici 2050. Si les objectifs fixés semblent à l’heure actuelle peu réalistes, la décarbonation des bâtiments reste néanmoins le potentiel le plus important (c.-à-d. plus de 50 %) pour atteindre les objectifs des accords de Paris et la stabilisation du changement climatique. La construction et l’opération de bâtiments a au moins la chance de devenir neutre ou négatif en émission carbone, ce que n’ont pas d’autres activités comme l’industrie ou le transport, en termes de cycle de vie. Pour se rapprocher des objectifs et espérer une réelle transition, une utilisation plus généralisée des technologies semble être une voie prometteuse.

Les nouvelles technologies au service de bâtiments plus responsables

En matière énergétique, l’apport des technologies est incontestable. Depuis plusieurs années maintenant, les bâtiments s’équipent d’outils de GTB (gestion technique du bâtiment) pour superviser et contrôler des services tels que le chauffage, la ventilation ou encore le conditionnement d’air, s’assurant qu’ils fonctionnent de la façon la plus efficace et la plus économique possible. En créant ces réseaux électriques intelligents (smart grids), la GTB s’est progressivement imposée comme la pierre angulaire des bâtiments intelligents et son interaction avec d’autres bâtiments, la smart city et les réseaux urbains, permettent d’envisager une décarbonation complète du parc de bâtiment. ]Avec la prolifération des objets connectés au sein des bâtiments, de plus en plus de données sont désormais disponibles, ouvrant la voie aux technologies d’intelligence artificielle. Les données recueillies servent à améliorer la compréhension et l’anticipation des usages des occupants, à automatiser certaines tâches, et, rapprochées d’autres données extérieures, à déployer des solutions pour gérer durablement les bâtiments : décarbonation, efficacité énergétique, conformité RSE des produits/fournisseurs, services de bien être des occupants, qualité de l’air, productivité du bâtiment (climatiser en fonction du taux d’occupation), coût du m2, contrats d’achat d’énergie, mix énergétique du scope 2 des émissions, économie circulaire, production solaire locale, chargement électrique de véhicules, etc. Seule l’IA permet l’apprentissage automatique et l’optimisation en prenant en compte autant de paramètres.

Dès qu’un paramètre évolue (changement de réservation de salle, météo des prochains jours, nouvel équipement de stockage d’énergie, évolution de l’occupation, chargement électrique de voitures dans les parkings, etc.), l’intelligence artificielle prend la relève pour adapter et optimiser les réglages au-dessus des algorithmes traditionnels. Une standardisation des données est évidemment importante (BIM, Brick), pour créer un jumeau numérique pérenne durant la vie du bâtiment.

Le bâtiment intelligent doit se muer et se généraliser

Si les technologies sont matures, le parc immobilier européen n’est quant à lui pas prêt à être équipé. Aujourd’hui, trop peu de bâtiments alignent tous les prérequis pour servir de plateforme à une offre de services d’intelligence artificielle. Seuls 1% des bâtiments en Europe sont en effet conçus pour être optimisés par l’IA et seulement 1% sont rénovés chaque année. La communauté européenne souhaite rapidement passer à 3 % de taux de rénovation, mais là, ce sont les ressources financières et humaines actuellement engagées qui ne sont pas forcément suffisantes pour suivre le rythme.

De la GTB à l’IA, il n’y a malheureusement pas qu’un pas. Pour avoir un bâtiment sain au niveau énergétique, il faut d’abord utiliser pleinement tous les algorithmes de GTB disponibles (démarrage optimal des installations – le premier algorithme GTB à apprentissage automatique développé il y a 50 ans, arrêt/marche de l’éclairage en fonction de l’occupation des locaux, gestion des stocks d’énergie), avant de pouvoir apporter une valeur ajoutée avec l’IA. Or seulement 10 % des bâtiments en Europe sont pleinement optimisés avec la GTB. Heureusement, le décret BACS prévoit de rendre obligatoire les systèmes d’automatisation et de contrôle GTB dans les bâtiments tertiaires d’ici le 1er janvier 2025. Si la transition s’effectue convenablement et qu’au moins 80 % des bâtiments sont optimisés grâce à la GTB, ce sont près de 20 % des bâtiments qui pourraient recourir aux technologies d’IA et réduire ainsi leur impact carbone. Soyons optimistes !