Glossary

Plain definitions of the terms we use around hackathons, bounties, and agentic AI. Each term has its own link.

Hackathon #: A time-boxed building competition, usually from a day to a few weeks, where teams ship a working project and submit it against a published judging rubric for prizes. Most are run by or for a sponsor whose tools the winning projects tend to use well.
CFP (Call for Proposals) #: A Call for Proposals is an open invitation to submit talks, projects, or entries to an event before a deadline. In a hackathon context it is the window during which teams register and declare what they intend to build.
Devpost #: A popular platform for hosting hackathons and managing submissions. Sponsors run events on Devpost, publish rules and prize tiers there, and collect project submissions (repo, demo video, and writeup) through it. Devfolio, ETHGlobal, DoraHacks, MLH, Unstop, HackerEarth, Kaggle, and Taikai are other platforms in the same space.
Google ADK (Agent Development Kit) #: Google's Agent Development Kit, a framework for building AI agents as structured systems of agents, orchestrators, and typed tools. It handles turn-taking, tool routing, and session state so builders can focus on the logic of each step. We used it to build Atlas.
Gemini #: Google's family of large language models. In an agent, Gemini serves as the reasoning engine for open-ended steps such as diagnosing a root cause or estimating impact, while cheaper deterministic code handles the numeric work around it.
BigQuery #: Google Cloud's serverless data warehouse for running SQL analytics over large datasets. In our Atlas project it is both the subject the agent monitors and the store where the agent writes its own run history.
MCP (Model Context Protocol) #: The Model Context Protocol, an open standard for connecting AI models and agents to external tools and data sources through a common interface. It lets an agent call a tool or read a resource without a bespoke integration for each one.
Agentic AI #: AI that does work rather than only answering. An agentic system plans, calls tools, acts on an environment, observes the result, and loops, often with limited human oversight. The test is whether it closes a real task loop end-to-end, not whether it can chat about one.
Autonomous agent #: A software agent that decides and acts on its own toward a goal, typically running a sense, decide, act, verify loop without a human driving each step. Well-designed ones keep autonomy legible: safe actions run automatically, risky ones are proposed for human approval. Atlas is an example.
Bug bounty #: A program that pays independent researchers for responsibly reporting valid security vulnerabilities in a company's software, scaled by severity. Scope and rules define what counts as a valid finding for that program.
Audit contest #: A time-boxed, competitive security review, common in smart-contract and web3 work, where many auditors examine the same codebase and share a prize pool based on the severity and uniqueness of the issues they surface.
Testnet #: A test network that mirrors a live blockchain but uses valueless tokens, so developers can deploy and exercise contracts without risking real funds. Hackathon submissions often ship on a testnet.
Judging rubric #: The set of weighted criteria judges use to score submissions, for example technical difficulty, design, impact, and presentation, each with a weight. We treat the rubric as the actual specification of what wins and build to the weights.
Prize stack #: The full structure of prizes at an event: a grand prize plus sponsor-specific track prizes, sometimes stackable so one project can win several. The shape of the stack reveals who the judges are and what the sponsors most want to see.

Read the FAQ → Read the blog →