Inspiration ## Inspiration
Research from Solidus Labs found that over 98% of tokens launched on platforms like Pump.fun show signs of being scams, rug pulls, or pump-and-dump schemes. The core problem isn't a lack of data — it's that ordinary people can't read Solidity smart contracts. They buy a token without knowing the contract was deliberately designed so they can never sell it again. That gap between "the information exists" and "a normal person can actually understand it" is what ATLAS was built to close.
What it does
ATLAS is a Blockchain Intelligence Engine that analyzes Solidity smart contracts and gives a plain-language risk verdict: SAFE, CAUTION, or DANGER. Instead of a single black-box score, it breaks risk down into five explainable dimensions — Ownership, Liquidity, Tax, Blacklist, and Honeypot patterns — and shows exactly which patterns in the contract triggered each score. The goal is explainability first: not just "this is risky," but "here is why."
How we built it
ATLAS runs on a modular pipeline: Loader validates the input, Scanner extracts the contract's structure, a Feature Extractor detects risk-relevant patterns, five independent risk-dimension modules score those patterns, a Risk Engine aggregates them into a final numeric verdict, and a Reasoner turns the findings into human-readable explanations. The scoring is fully deterministic and reproducible — the same contract always produces the same result, and every reasoning line can be traced back to a specific pattern actually found in the code. The frontend is built with Streamlit. The reasoning layer is architected to use Gemini when an API key is available, with a transparent rule-based fallback when it isn't — the app always tells you honestly which mode it's running in, instead of pretending.
Challenges we ran into
This part of the story matters to me: I built this entire project from my phone. No laptop, no desktop — just a phone, GitHub Codespaces, and a lot of persistence. Copy-pasting long code blocks kept getting corrupted or cut off mid-transfer. Mobile terminals didn't behave the way tutorials assumed. At one point GitHub Codespaces hit its monthly usage limit right as I was trying to finish integration. I'm 18, and a lot of this process was genuinely me figuring things out step by step, hitting a wall, and finding a workaround — over and over — rather than already knowing the "right" way to do it.
Accomplishments that we're proud of
Getting a real, testable backend pipeline working end-to-end — entirely from a phone — is the accomplishment I'm most proud of. Not a mockup, not hardcoded demo values: an engine that actually parses contract source code and produces a different, reproducible score depending on what's really in the contract.
What we learned
I learned that shipping something honest is harder, and more valuable, than shipping something that just looks impressive. It would have been easy to fake a polished result. It was much harder — and more worth doing — to build something where every number on screen can actually be explained.
What's next for ATLAS
Connecting the Gemini reasoning layer live, adding real on-chain data lookups via RPC/explorer APIs for direct address analysis (not just pasted source code), and validating the scoring weights against a real dataset of known honeypot and safe contracts instead of heuristic judgment alone.
