Inspiration
The Web3 ecosystem suffers from a massive usability and security gap: the point of signature. Every day, users are prompted to sign raw hex calldata or Base58 instruction data that is completely unreadable to the human eye. This blind-signing relies entirely on trust, leading directly to devastating wallet drainers, infinite approval exploits, and lost funds. We wanted to eliminate this friction by bridging the gap between complex blockchain machine code and clear human intent, empowering users to actually understand what they are agreeing to before they hit "Approve."
What it does
Readable Receipt is an AI-powered transaction analysis engine. It takes raw, unreadable blockchain payloads—whether it's EVM hex calldata, Substrate/Midnight extrinsics, or Solana Base58 instruction data—and translates it into a clear, human-readable security summary in milliseconds.
Instead of staring at a wall of alphanumeric characters, users receive:
Risk Assessment: A definitive action badge (Ignore, Review, or Escalate).
Human Summary: A plain-English explanation of the transaction's exact intent (e.g., "You are granting an unlimited token approval to this smart contract").
Asset Impact: A precise breakdown of which tokens and amounts are entering or leaving the wallet.
How we built it
Frontend Interface: Built utilizing a modern, responsive UI css. We implemented strict client-side Regex validation to immediately drop invalid payloads based on the selected network (e.g., ensuring even-length hex strings for EVM/Midnight, and Base58 compliance for Solana) to save compute and prevent errors.
Edge Compute: We deployed our backend logic on Cloudflare Workers to ensure ultra-low latency requests, processing the payloads as close to the user as possible.
AI Security Engine: We integrated the Gemini AI API within the worker to act as the decryption layer. By passing the sender, target program ID, and validated calldata, the AI cross-references the data and outputs a strictly formatted JSON response mapping the transaction's intent and risk level.
Challenges we ran into\
One of the most significant hurdles was standardizing the AI's output. Generative AI naturally wants to be conversational, but our frontend required a highly predictable JSON schema to drive the UI state (colors, badges, and asset lists). We spent considerable time iterating on our prompt engineering and system instructions to force the AI to return clean, parsable JSON without markdown wrapping. Additionally, building unified validation logic that could gracefully handle the stark differences between EVM hexadecimal structures and Solana's Base58 encoding required careful regex implementation.
Accomplishments that we're proud of
We are incredibly proud to have built a highly functional, cross-chain security tool within the intense time constraints of the MLH Midnight Hackathon. Successfully connecting a seamless, pixel-perfect frontend to a Cloudflare Worker that dynamically queries an LLM and returns formatted data in milliseconds is a major technical win. We proved that AI can be used not just for chat, but as a real-time security infrastructure layer for Web3.
What we learned
This project pushed us deep into the architecture of transaction payloads across completely different ecosystems. We learned the nuances of Solana's instruction data versus standard EVM calldata, and how to effectively structure edge functions using Cloudflare Workers. Furthermore, we leveled up our prompt engineering skills, learning how to tightly constrain AI models for deterministic, programmatic data extraction rather than open-ended generation.
What's next for Readable Receipt
The immediate next step is taking the core engine out of the web browser and packaging it directly into a browser extension. By automatically intercepting "window.ethereum.request" and "Solana" wallet adapter events, Readable Receipt will sit silently in the background and pop up its analysis before the wallet's native signature screen even appears. We also plan to expand our "Threat Gallery" by implementing non-AI heuristic checks (like known malicious contract registries) to flag threats even faster before they hit the AI engine.
Log in or sign up for Devpost to join the conversation.