luisalarcondelalastra_63916
Project URL
Code URL
Section
Technical Architecture Doc
Team bio
Luis Alarcón
The company is led by, Luis Alarcón, the CEO, (linkedin - https://www.linkedin.com/in/luis-alarc%C3%B3n-de-la-lastra-810113122/) is an AI Engineer, full-stack developer who also holds Masters in Data Science, Software Development and ir currently studying a Masters in Complex Problem Solving. Providing a holistic view of systems and architecture development and data-centered iteration in order to constantly keep improving systems. Luis also holds a Master's degree in Electronic Music Composition and a Business and Management degree. Missio IA combines a strong base of engineers with a network of experts from various fields. This multidisciplinary team addresses problem-solving across different sectors, providing tailored and strategic solutions that enhance organizational efficiency and growth through advanced AI applications.
Estela Falgas
Estela is a Mathematician, with Masters in Quantitative Computing, Philosophy of Science and bootcamp in Data Science. (linkedin - https://www.linkedin.com/in/estela-falgas-168852303/) She leads the research area of the company, with special focus on Databases and Data Science, also leading the iteration department, where she uses the scientific method to constantly iterate and improve agentic systems.
Ignacio Satrustegui
Ignacio holds an MBA and a degree in Economics, while still coding and being highly involved on the technical side, he leads the sales side of the company and is constantly in contact with customers and other stakeholders. (linkedin - https://www.linkedin.com/in/ignacio-satr%C3%BAstegui-006873195/) This allows us to capture user feedback and keep improving the solutions we build.
A main area of interest right now for the team is the blockchain and its fusion with AI. The Soroban ecosystem seems like the correct place for us as we want to know and properly understand it deeply, get a true feel of what is possible, help it grow, and grow with it. Definitely continue with our mission of shedding light in the unknown.
Pitch Deck URL
Project Categories
Product Type [If Other Developer Tooling]
Section
Requested Budget in USD ($)
Section
Public Entity Name
Entity Description
Progress on Previous (Awarded) Submissions
- We have done 2 iterations of the product and have a working prototype. We are commited to the project, and want to demonstrate it. We’ve researched the rest of AI solutions, and after receiving funding, none of them work (excluding the Stellar GPT which is not tailored to build Smart Contracts) while we have a working prototype pre-funding. The prototype is designed specifically for liquidity pool contracts as a proof of statement, but with funding we would have the needed support for dedicating the time and resources needed to work on expanding up to maximal generalisation, with capability to build for any type of new use case.
Iteration 2:
We have a chat interface to give all needed info to the system to get started with contract building (pending is to also reflect on the optimal interface for this task). The resulting zip with all contract files will be sent by email to the developer (os SCF reviewer). Here is a demo video of iteration number 2: https://drive.google.com/drive/folders/1_YfJ9zZ5IDnr77XOHM5kpiLzO_5wmafb?usp=sharing
Here is a link to test it out live whenever needed: https://sorodemogit-fem5yo2zby6yjqwgxztvgv.streamlit.app/
-
We are experts at using AI agentic flows to solve complex problems. Agentic flows are a structured series of actions and operations managed and executed by AI independent agents. These agents act as autonomous entities which are in charge of solving a very specific aspect of the problem/solution that is being addressed. This approach allows us to break down a very complex task such as building a Soroban Smart Contract assistant into a lot of smaller tasks that are much easier to handle individually.
An example of this is having a specific agent that is solely in charge of understanding the problem a developer is trying to solve with a smart contract, then, a specific agent that converts this into a comprehensible in-depth problem statement, then followed by another agent that compares that problem to a vectorized database of problems that have been previously solved by using smart contracts.
You can find a full description of the whole flow and all the independent agents in the technical architecture doc.
-
Secondly, given that we are LLM-agnostic (we have worked with +10 different models for different solutions we have previously built) and that we work with an agentic flow, we can test and iterate the system with different specific models for each of the independent functions.
For example, it might be that GPT-4o is the best model to understand what the user wants to solve and define a comprehensible problem statement, while Llama 3 (with Groq) might be the best at actually building the different code files for its speed, while Codepal is the best at reviewing and debugging the code. Our plan is to try hundreds of combinations in order to finetune the system.
- Also, the timing is perfect to build a Soroban AI assistant. 4.1 Soroban being live now on Mainnet means that there is a wide variety of functional contracts which have been tested and that we can vectorize so that one of the agents in the flow can have full knowledge of all the examples of functional contracts that are available. This ensures that the “smart-contract architect” agent has the capability of proposing a logical smart contract architecture to solve the problem statement that has been given to him. 4.2 Soroban being live on mainnet means that once our product is live it can help in the onboarding of new developers that want to build smart contracts, as by utilizing a very simple front-end and Natural language, they will be able to go from a problem statement to a robust foundation of code that aims to solve the issue they are trying to address. This makes a lot of sense together with our “Documentation Agent”, as it will fully document every step of the process and explain each part of the code so that the developer can understand the logic behind each smart contract element. 4.3 There are live testing tools at the moment such as Okashi.dev which allow for the testing of the outcome of our product very easily, allowing us to quickly test the outcomes of our system and keep iterating and improving. This ensures that we can quickly measure the impact of our changes on the final outcome. Moreover, once our product is fully live, developers will be able to easily deploy their contracts and detect errors… by using these same tools.
Deliverable and Budget: High budget but deliverables are well-detailed.
We understand it is a high budget, this is because the project is very time intensive and will require hundreds of hours of AI Engineers testing different iterations and combinations of the agentic flow, different models for each function... However, after internal consideration, we have decided to cut the budget by 15% which we will cover with resources of our own.
Furthermore, we have also added more details to the deliverables to further explain our action plan and how we plan to succeed step-by-step.
Soroban is currently experiencing a significant influx of developers eager to build within the Stellar ecosystem. As a result, many of these developers in the coming months and years will likely be new to Soroban and Rust. To support Soroban developers in conceptualizing and writing smart contracts, we are developing an AI-powered smart contract generation assistant. This assistant will possess agentic capabilities, enabling it to gather the necessary information and deliver a smart contract project ready for review, testing and iteration on the Testnet.
It is key to highlight that as the Soroban ecosystem develops, more data will be available, and together with the exponential improvements LLMs are experiencing, the Soroban AIssistant will also keep improving exponentially. This will widely help the ecosystem grow and increase the accessibility to developing on Soroban.
Product & Architecture
Soroban is currently experiencing a significant influx of developers eager to build within the Stellar ecosystem. As a result, many of these developers in the coming months and years will likely be new to Soroban and Rust. To support Soroban developers in conceptualizing and writing smart contracts, we are developing an AI-powered smart contract generation assistant. This assistant will possess agentic capabilities, enabling it to gather the necessary information and deliver a smart contract project ready for review, testing and iteration on the Testnet.
We have already developed a proof-of-concept Soroban specific system which successfully outputs complete robust Soroban smart contracts with little errors. You can find a video of how it works in the Drive.
The code generated by the assistant should be viewed not as a final product but as a robust foundation for the project.
This tool will significantly accelerate the development process. The idea is to add another catalyst in the development workflow that further “Sorobinices” the Stellar “Abacus”, code-included as much as batteries included. In a rapidly growing system still in its infancy, we aim to provide a multiplier that effectively drives feedback and fosters growth.
The Soroban AIssistant is a web-app that runs in the background on an AI agentic system specifically designed and tailored to the creation of Soroban Smart Contracts. It will allow Soroban developers to describe in depth the problem they are trying to solve in a conversational way and the assistant will ask all necessary questions to collect the information it needs to proceed.
By compiling and vectorizing all available information on Soroban (including Stellar’s documentation, APIs, Github code libraries…) as well as the Smart contracts that have been deployed on Github, the whole model is solely focused on being a tool Soroban developers can use to accelerate the conceptualization and creation of smart contracts.
The agentic flow consists of 3 main stages. We will provide a high-level overview of the whole process here, but you can find the full detail and deep dive in the technical architecture, where you will be able to find all details on the different agents and how the system interacts given that it is a deep-tech project.
-
Understanding Phase and Initial proposal creation: The first part of the flow is aimed at fully understanding what the user is trying to solve or build gathering Gather comprehensive user inputs and converting them into a coherent problem statement. Then, considering all available options to tackle it and merge all user inputs together with the vectorized database of Soroban, Stellar and Rust knowledge to conceptualize a potential solution.
-
Refinement phase: Propose and refine an initial design using specialized AI modules. This phase focuses on integrating expert inputs and iterative improvements to ensure the design is robust and ready for development.
- Building the Smart Contract: Construct, review, and document the smart contract based on the refined design. Output zip folder with project files and folders for testing.
The user will recieve a zip file with all project files and a .md document with comments and explanations, fully documenting every step of the process, decisions made and so they can navigate the code much easier.
Deployment and Testing: After this, the developer can directly upload and test the smart contracts on the testnet and by utilizing a tool such as the Okashi.dev playground, they can start building on top of that foundation. It is key to highlight that as the Soroban ecosystem develops, more data will be available, and together with the exponential improvements LLMs are experiencing, the Soroban AIssistant will also keep improving exponentially. This will widely help the ecosystem grow and increase the accessibility to developing on Soroban.
We have many ideas of improvement for the coming moths such as outputting also the server side Python and Javascript files, chatting with the docs, and possibly giving access to more specific agents that fulfil a single role very well. For example a Security auditor or an oracle creator.
Video URL
Deliverables List
The main objective of the Activation Award is to have a live MVP which developers can interact with and test as soon as possible. This will allow us to keep iterating and improving both the individual AI agents, as well as the product itself and the overall smart contract conceptualizing and writing capabilities.
The best way to measure the successful completion of the deliverables will be to directly utilize the platform, generate a smart contract and test it on Okashi.dev for example. We will also provide links to all the code.
The biggest cost driver of the development of the project will be the salaries of AI Engineers and Software Developers, as constantly iterating the whole model and all its components is a key element to ensure success, but will be highly time intensive.
Deliverable 1:
- Data recollection and vectorization of both code and written available Soroban Information.
- RAG Strategy Testing.
Although we have already started doing so, we are vectorizing all Stellar and Soroban information including the documentation, GitHub Smart contract examples by the Stellar Org, smart contracts developed by individuals… All of these vectorized databases will be used in different ways by the different agents.
Integrated in the cost is the testing and research into unlocking the optimal RAG strategy for this task.
Estimated time: 3 weeks after award is received
How to measure completion: Improvement in agent performance and smart contract generated quality. Our goal is to be able to test smart contracts in Okashi without errors when we launch our pilot so the efficacy there will be in part thanks to RAG R&D research. We'll also provide RAG metrics evolution over time.
Budget: $5.000
Deliverable 2:
Development of all the specific agent assistants involved in the system (in depth detail about each one can be found in the technical architecture document). Including:
- User Input Gathering AI
- Problem Proposer AI
- Initial Design AI
- Specialist AIs: Horizon API and RPC Server Expert: Focuses on Horizon API and RPC server interactions.
- Dynamic Area Experts: RPC Providers, Oracles Expert, Cross Contract Expert, Security Expert
- Iterative Improvement and Research
- Improved Design AI
- Builder AI
- Final Review AI
- Documentation AI
- Compilation and Packaging.
This includes all types of testing with different of models and experimenting with DSPy, and other frameworks to get maximum performance.
How to measure completion: We will provide a link to the code of all of these agents so they can be reviewed. Also smart contract output quality.
Estimated time: 6-8 weeks after award is received.
Budget: $21.250
Deliverable 3: System Behavior Testing & IteratingDifferent test and architectures will be developed in the path to optimal performance. It is here where we'll iteratively improve test behaviour. Part of the cost in in also finding a way to self-optimize this part as we did with individual agent nodes and the use of DPSy. Lot's of manual testing, improvement supervision and human direction will be needed in this part without proper self-optimizing mechanisms in place. Part of the cost is in developing self-improvement mechanisms.
How to measure completion: Improvement in agent performance and smart contract generated quality. Our goal is to be able to test smart contracts in Okashi without errors when we launch our pilot so the efficacy there will be in part thanks to flow engineering and R&D. Final system architecture will be seen in code and documentation with different iterations mentioned.
Estimated time: 6-8 weeks after award is received
Budget: $11.250
Deliverable 4: Web-App Platform Building MVP
The development of an MVP web-app that will allow for the interaction with the Soroban AIssistan.
How to measure completion: The platform will be live and we will provide the URL of the live product.
Estimated time: 6 weeks after award is received
Budget: $1.000
Deliverable 5: Smart Contract Security R&D
Although we'll work with smart contract experts for the whole scope of the project, it is specially here where we are most focused on getting it right in order to manage align with Stellar values ASAP.
How to measure completion: The platform will be live and we will provide the URL of the live product.
Estimated time: 6 weeks after award is received
Budget: $4.000
Total Expected Roadmap
In order to have the product be self-sustainable, we have determined three time horizons, activation award, community award and post-community award (where we will rely on the product’s revenue and our own funds to keep the product growing).
Activation Award: The objective of the Activation Award is to have a functional MVP Platform which Soroban developers can start utilizing and testing 1.5 to 2 months after receiving the award.
Community Award: Once the MVP is live, it will allow us to quickly gather feedback, collect data points and start building a much more robust Soroban Assistant, constantly increasing the value to the community. We will also start putting a lot more resources towards the usability of the platform (Web-app), prioritizing user experience to accelerate the adoption of our product together with further improvement to the system.
After this, we will start transforming the Soroban AIssistan into a SaaS, ensuring the long term sustainability of the project.
Growth and continuous improvement: As mentioned at the start of the proposal, with Soroban being at it’s starting stage and the exponential improvements in LLMs, we aim to constantly keep developing the Soroban AIssistan, so it can constantly become more helpful and more relevant in the Stellar ecosystem.