1. Overview

<aside>

DevAsign is an AI platform that automates bounty payouts via smart merge & escrow smart contract, enabling OSS maintainers focus on building rather than admin overhead.

</aside>

https://www.figma.com/board/XMe9YomhxgNqBZJZcyXomI/devasign-|-user-flow--MVP-?node-id=0-1&t=RT8R15QduYeFOSMz-1

DevAsign uses a fine-tuned Anthropic Opus-4 with a LangChain-based RAG pipeline. On each PR, it fetches context, runs checks, and scores merge confidence (0–100). It only recommends by default — auto-merge is opt-in and guarded by repo rules and safety modes specified by the maintainers.

2. AI System Architecture

Layer What we use Implementation Notes
Base model Anthropic Opus-4 Fine-tuned
Retrieval index Vector database storing embeddings of: • Entire repo at HEAD • Last 12 months of closed issues and their corresponding pull requests • Current YAML/JSON rule file
RAG assembler Pipeline that: • Accepts the incoming PR difference • Retrieves top-K relevant code chunks, past PRs, and the project’s rules Built with LangChain
Rule engine Plain YAML/JSON schema

What the AI bot actually does

3. API Layer


4. Core Microservices

4.1 PR Merge Service (Node.js + TypeScript)