diff --git a/applications/inkjet.md b/applications/inkjet.md new file mode 100644 index 00000000000..1695d6c0444 --- /dev/null +++ b/applications/inkjet.md @@ -0,0 +1,259 @@ +# ink!jet + +- **Team Name:** Parallel Polis +- **Payment Address:** Fiat USD 17.12.2023, 20:14 PST +- **[Level](https://github.com/w3f/Grants-Program/tree/master#level_slider-levels):** 2 + +## Project Overview :page_facing_up: + +This application is *not* in response to an RFP nor a follow-up grant. + +### Overview + +**One liner:** Platform for Generative AI Assisted Smart Contract Development with Retrieval Augmented Generation + +`Ink!jet` is a platform designed to use augmented generative AI with the intent of improving the development lifecycle of `ink!` smart contracts within the Polkadot ecosystem. Recognizing the technically intricate nature of smart contracts and the high level of expertise they demand, our project aims to simplify these complexities, thereby democratizing the creation process. + +Existing generative AI have limitations in both the amount of `Rust` and `ink!` code in their training data. Our platform uses a retrieval-augmented generation pipeline with datasets of existing `ink!` smart contracts to bridge this knowledge gap. Injecting vectorstore retrieved code into prompts, this system utilizes in-context learning to improve response quality. Our goal is to enhance the productivity of existing developers through bootstrapping and assisted code interation, while simultaneously lowering the barrier of entry for new developers. + +Our team's motivation for the project is twofold. First, we are supporters of decentralized technology and its potential to redefine the digital landscape. We believe that the future of blockchain depends on cross-chain interoperability, which places Polkadot at the center of this emerging need. Second, we are deeply intrigued by the potential of generative AI in software engineering. Our aim is to push the boundaries of what these AI models can achieve in novel contexts, such as the `ink!` programming language, and to explore solutions that empower developers to build within the decentralized ecosystem. + +### Project Details + +#### UI Mockup: + +We are opting for a simplistic UI for ease of use. The tool will be a single-page application, where the left side of the page is an in-browser IDE for the `ink!` smart contract being worked on. The right side is a Chat History where the user can interact with the LLM. Code generations in responses will be automatically populated in the left IDE. + +We provide a dropdown menu of code templates to choose from. The History button will keep a edit history of the smart contract if the user needs to revert to a previous version. The Analyze button will break the code down into chunks (such as its individual functions) and provide feedback for each chunk. The feedback will replace the Chat History, and can be closed once it is read to resume editing. + +Users are provided parameters to adjust with their prompts, and can modify the temperature and number of retrieval documents from the vectorstore. + +![ui-mockup](https://jyu.llc/inkjet_mockupui.png) + +#### Data Model / Architecture + +![architecture](https://jyu.llc/inkjet_arch.png) + +#### Stack + +We will be using the following technologies: +- Python for RAG-LLM pipeline + - LlamaIndex and LangChain libraries for data loading, processing, embedding + - LlamaIndex and LangChain libraries for vectorstore retrieval and LLM interaction with retrieval results +- Pinecone for Vectorstore +- OpenAI text-embedding-ada-002 Model for Embeddings +- OpenAI for LLM (GPT-4-32k, GPT-4-1106-preview) +- ink!/Rust for Smart Contracts +- React.js for Front-End Application +- Docker for Containerization +- Vercel for Hosting + +#### Relevant Work + +Please see Development Status section. + +#### What this project is NOT + +We are not providing a for-profit product, and all usage will be free to any developers. The system and datasets will also be public, so developers can choose to host their own version of the system, with modifications if they please. + +We are not providing full automation nor replacement for existing developers, this tool is designed to enhance the developement cycle and increase efficiency. + +We are not claiming that this system is without fault. Though the system is aimed at mitigating errors and vulnerabilities, there is a degree of inherent randomness when using LLMs for code generation. We will provide stringent disclaimers and advice to users to rigorously test their code before deployment, and advocate for contract auditing. + +### Ecosystem Fit + +This project fits into the ecosystem as a developer tool. It is aimed at improving the smart contract development cycle through bootstrapping, assisted coding, and iterative feedback. + +The target audience are existing smart contract developers and those looking to start writing smart contracts. + +As the programming language ink! is built on top of Rust, there exists some barriers of entry, for both Rust and ink!. + +Rust is a low level language, requiring manual management of memory and pointers. While the language has steady increases in adoption and high ratings, it remains challenging for those coming from other languages and especially new developers. In addition to this complexity, the ink! programming language introduces a handful of macros and does not rely on the standard `std` library. Moreover, managing environment variables are done with a different crate and the code is compiled into WASM instead of machine code. + +Acknowledging these technical barriers, our tool aims to facilitate an easier transition towards developing in ink! through guided development. Those who already know Rust can easily step into ink! development, and have the differences in syntax explained. Those who already know ink! will be able to save time writing boilerplate and refining work-in-progress contracts. Users are continuously provided feedback on their code, which saves time spent searching documentation or posting on forums. + +At this point in time, current LLMs provide limited assistance due to the scarcity of Rust and specifically ink! code in their training data, and we aim to bridge this gap through our approach. + +To the best of our knowledge, there are no existing projects that are similar at this time. + +## Team :busts_in_silhouette: + +### Team members + +Jeffy Yu + +Maximilian Huber + +Kevin Tang + +### Contact + +- **Contact Name:** Jeffy Yu +- **Contact Email:** jeff.yu@parallelpolis.llc +- **Website:** jyu.llc + +### Legal Structure + +- **Registered Address:** 656 W Nopal Ave Mesa, Arizona 85210 +- **Registered Legal Entity:** Parallel Polis + +### Team's Experience + +We are a research collective conducting open source research on areas in blockchain, machine learning, artificial intelligence, and general software engineering. + +Our team has an academic background in computer science, and work experience in blockchain development, machine learning, and full stack development. We are looking to bring our skills towards a project that can improve accessibility of the Polkadot developer ecosystem. + +Jeffy Yu - Former full stack blockchain engineer at Paystand, a B2B payments company anchoring invoice records on the Ethereum blockchain. A member of the R&D team at Paystand, built a payment network in Rust supported by the Lightning Network to migrate invoicing towards decentralized solutions. Former full stack engineer at Lirn, a decentralized learning platform utilizing ERC-1155 tokens as educational certificates. Research Scientist at Parallel Polis, performing open-sourced technical studies on blockchain and artificial intelligence. Recently published work on Lightning Network topographies for closed payment systems, accessible [here](https://scholar.google.com/citations?user=zy7ZWqYAAAAJ&hl=en). + +Maximilian Huber: Former full stack engineer at CollX Inc., experienced in building RAG retrieval systems and deploying generative AI for specific use cases. Experience building user-facing LLM-enabled applications. + +Kevin Tang: Full stack engineer with primary experience in JavaScript and React.js, as well as experience in iOS-development at Priceline.com. Experience developing and maintaining systems in active deployment. + +### Team Code Repos + +- [https://github.com/yu-jeffy/audit.me](https://github.com/yu-jeffy/audit.me) +- [https://github.com/yu-jeffy/GreedLlama](https://github.com/yu-jeffy/GreedLlama) +- [https://github.com/yu-jeffy/PyArb](https://github.com/yu-jeffy/PyArb) +- [https://github.com/yu-jeffy/gpt-btc](https://github.com/yu-jeffy/gpt-btc) + +### Team Github Accounts + +- [https://github.com/yu-jeffy/](https://github.com/yu-jeffy/) +- [https://github.com/MaxHuber88](https://github.com/MaxHuber88) +- [https://github.com/tangk01](https://github.com/tangk01) + +### Team LinkedIn Profiles + +- [https://www.linkedin.com/in/jeffyyu/](https://www.linkedin.com/in/jeffyyu/) +- [https://www.linkedin.com/in/max-huber88/](https://www.linkedin.com/in/max-huber88/) +- [https://www.linkedin.com/in/tangk01/](https://www.linkedin.com/in/tangk01/) + + +## Development Status :open_book: + +We have conducted extensive research on existing work and the overall feasibility of the project. Here are our findings: + +To our knowledge at this time, there are currently three publications assessing the efficacy of LLMs for smart contract auto-completion and generation. +- [https://arxiv.org/abs/2308.02955](https://arxiv.org/abs/2308.02955) + +This publication uses the base model without additional learning, and finds limited success, with security bugs and errors being frequent. We look to address this by providing additional context to prompts through our RAG system, which allows GPT-4 to perform in-context learning. + +- [https://arxiv.org/abs/2311.10388](https://arxiv.org/abs/2311.10388) + +This publication utilizes in-context learning with a retrieval system, similar to our approach. They focus solely on generating code comments, not on code itself. They find improved efficacy, which is promising as the design is similar to our system. + +However, their design is distinct from ours as they utilize CodeBert to create the embeddings, which has no Rust code in its training data. This makes it incompatible with ink! smart contracts. They also utilize an older model of GPT, GPT-3.5-turbo, which has been significantly outperformed by GPT-4 on coding benchmarks. + +We aim to create an alternative system and while replicating their efficacy improvements, using an embedding model that supports Rust and a more recent GPT model, and extend functionality beyond code comment generation to smart contract code generation. + +- [https://arxiv.org/abs/2309.09826](https://arxiv.org/abs/2309.09826) + +This publication fine-tunes a GPT-J model on 2 million smart contracts, and tests if performance increases on writing code without security issues. While their methodology involved directly fine-tuning the model and updating the weights, and ours focuses on adding context through retrieval, the underlying concept of providing learning examples to improve performance remains similar. After the original fine-tuning, they find that insecure code was an issue in up to 70% of generations. After additional fine-tuning on vulnerable examples with vulnerability-constrained decoding, they were able to avoid insecure code generation up to 67% of the time. We have taken these findings into consideration, and will implement labeled vulnerable examples in our dataset. This allows our system to identify vulnerabilities if present in the user's smart contract, and also avoid generation of them in code responses. + +In terms of related work, we have [ongoing work](https://github.com/yu-jeffy/audit.me) studying the efficacy of vulnerability testing through RAG integrated LLMs on Ethereum Solidity smart contracts. In this study, we built a LangChain RAG-LLM pipeline, and created a vectorstore of 830 vulnerable Solidity smart contracts for retrieval. Results were promising, showing a 1.5x increased efficacy compared to [current literature](https://arxiv.org/abs/2306.12338). Using this as a proof of concept of RAG-LLM with smart contract data, we look to rebuild the pipeline towards the use case of authoring ink! smart contracts. We will construct a new pipeline from scratch, utilize LlamaIndex alongside LangChain, change the data processing and embedding methodology, and implement a much broader ink! smart contract dataset. + +## Development Roadmap :nut_and_bolt: + +### Overview + +- **Total Estimated Duration:** 5 months +- **Full-Time Equivalent (FTE):** 3 FTE +- **Total Costs:** 30,000 USD + +### Milestone 1 — Prototype, Initial RAG System and Data Pipeline + +- **Estimated Duration:** 1 month +- **FTE:** 3 +- **Costs:** 6,000 USD + +| Number | Deliverable | Specification | +| -------: | ----------- | ------------- | +| **0a.** | License | Open-sourced under Apache 2.0. | +| **0b.** | Documentation | Code comments. Documentation for the prototype architecture and the setup process, explaining how a user can run a local instance of the prototype RAG system with our initial data. | +| **0c.** | Testing Guide | Unit tests will be provided for the prototype, with a testing guide explaining how to run them. | +| **0d.** | Docker | We will include a `Dockerfile` that enables easy deployment and testing of the RAG system. | +| 1. | Initial Prototype | Development of a basic LlamaIndex RAG system prototype integrated with `GPT-4`, using sentence embeddings. User can interact with the pipeline through the command line, interfacing with `GPT-4` with fetched documents from `Pinecone`| +| 2. | Data Collection | Collection of a small set of `ink!` smart contracts for initial embedding and retrieval testing. Smart contracts will be converted from `.rs` files to `JSON`, with identifying metadata. | +| 3. | Loading and Embedding Pipeline | Pipeline for loading in `ink!` smart contracts as `JSON` files, and generating embeddings to build the initial vector database. | +| 4. | Vector Database | `Pinecone` database will be used to store embeddings from processed `ink!` smart contracts. | + +### Milestone 2 — Smart Contract Dataset, Embeddings for Vector Database + +- **Estimated Duration:** 1 month +- **FTE:** 3 +- **Costs:** 6,000 USD + +| Number | Deliverable | Specification | +| -------: | ----------- | ------------- | +| **0a.** | License | Open-sourced under Apache 2.0. | +| **0b.** | Documentation | Code comments. Documentation on the vectorization of newly collected smart contracts and updating the vector database. | +| **0c.** | Testing Guide | Unit tests for changes to loading and vectorization. Testing guide included for running these tests. | +| **0d.** | Docker | `Dockerfile` updated to reflect any changes in RAG system and data pipeline deployment. | +| 1. | Data Expansion | Collection, categorization, and metadata tagging of `ink!` smart contracts to complete the dataset. Rigorous code commentation in the smart contracts to provide semantic context before vectorization. | +| 2. | Embedding Model | Replace default sentence embedding. Implement OpenAI's `text-embedding-ada-002` model, compatible with natural language and code processing. | +| 3. | Vector Database | Update the `Pinecone` vector database to house the code-based embeddings of the complete dataset. | + + +### Milestone 3 — RAG System Integration + +- **Estimated Duration:** 1 month +- **FTE:** 3 +- **Costs:** 6,000 USD + +| Number | Deliverable | Specification | +| -------: | ----------- | ------------- | +| **0a.** | License | Open-sourced under Apache 2.0. | +| **0b.** | Documentation | Code comments. Documentation provided for the integration process, and usage instructions provided for the updated RAG system. | +| **0c.** | Testing Guide | Unit tests updated to reflect changes in RAG system. Guide for testing the updated RAG system. | +| **0d.** | Docker | `Dockerfile` updated with any changes in deployment. | +| 1. | RAG Integration | Integration of the new dataset and vector database with the prototype RAG system for improved retrieval. Modifiable parameters for queries, including temperature, top-k documents in retrieval, and token context size. | +| 2. | Query Handling | Updated query parsing and processing using new embedding model to improve smart contract code retrieval. Prompt rewriting of plaintext in user request to improve semantic search. | +| 3. | System Testing | Testing of RAG retrieval functionality with the updated dataset. Testing of output quality of RAG system. | + +### Milestone 4 — UI/UX Development and Front-End Implementation + +- **Estimated Duration:** 1 month +- **FTE:** 3 +- **Costs:** 6,000 USD + +| Number | Deliverable | Specification | +| -------: | ----------- | ------------- | +| **0a.** | License | Open-sourced under Apache 2.0. | +| **0b.** | Documentation | Code comments. UI/UX design documentation and front-end codebase documentation, including setup and deployment procedures. | +| **0c.** | Testing Guide | Guide on how to run tests on the `React.js` application. | +| **0d.** | Docker | Updated `Dockerfile` to include the `React.js` application and associated dependencies. | +| 1. | UI/UX Design | Design and development of user interface for the RAG system. | +| 2. | React.js Application | Development of `React.js` front-end for the RAG system. Code implementation of the UI/UX design. | +| 3. | Front-End Testing | Comprehensive testing to ensure the front-end application is responsive and stable. | + +### Milestone 5 — Additional Features + +- **Estimated Duration:** 1 month +- **FTE:** 3 +- **Costs:** 6,000 USD + +| Number | Deliverable | Specification | +| -------: | ----------- | ------------- | +| **0a.** | License | Open-sourced under Apache 2.0. | +| **0b.** | Documentation | Code comments. Documentation of additional features, including use cases and integration into the existing system. | +| **0c.** | Testing Guide | Updated testing guide that includes all features of the application. | +| **0d.** | Docker | Finalized `Dockerfile` for the complete system with all features included. | +| **0e.** | Article | We will publish an **article** that showcases the development process, application use cases, application demo, and usage of the grant throughout the project cycle. | +| 1. | UI/UX Updates | Updated `React.js` components to serve additional features in the application. | +| 2. | Templates | Development of a template system for users to easily bootstrap building smart contracts. User can pick from a list of common use cases, and be provided a template smart contract to start with. | +| 3. | Smart Contract Analysis | A feature to break down smart contracts into chunks, scanning each for functionality and errors, and suggest improvements or fixes. Contract will be broken down into parts, such as its functions, and each will be analyzed by the LLM. | +| 4. | Final Application Testing | System-wide testing of all features, ensuring full integration and operational stability. | + + +## Future Plans + +- We plan to set aside a significant portion of the grant funding towards ongoing OpenAI API costs and other hosting costs. +- We will secure additional funding as needed for these ongoing costs if there is substantial demand on the platform. +- We intend to maintain the platform while it is deployed, fixing any bugs or issues users find. +- We look to add additional features to the platform, including Solang support and contract deployment testing capabilities. +- We look to continuously improve the vectorstore, through gathering of more smart contract examples and vulnerability avoidance. +- We look to improve the vector search, such as optimizing semantic relations with code annotations in specific formats. + +## Additional Information :heavy_plus_sign: + +**How did you hear about the Grants Program?** Web3 Foundation Website