SNEHA C

Ripple - the autonomous oncall agent crew with a FalkorDB Brain, built for the Graph Hacks hackathon

Graph Hacks/Oct 2026

When something breaks in production, the person on call is under pressure and has to answer questions that no single dashboard can. What depends on the failing service? Who has worked on it recently? Did we see this before, and did the fix we used last time actually work? The answers live in different places: metrics tools, pull requests, tickets, chat threads, and the memory of whoever was on call last. Today’s AI agents don’t fix this. They see pieces of the picture, forget earlier incidents, and can’t explain why they reached a conclusion, so engineers can’t trust them when it matters. Ripple is an on-call agent crew that uses FalkorDB as its context layer. Services, dependencies, people, pull requests, deployments, past incidents, runbooks, and decisions are all nodes and edges in one graph. When an alert arrives, Ripple opens an incident and a team of agents works on it, streaming each step live. The graph decides what the agents do. Triage walks the dependency graph to size the blast radius. The investigator builds competing hypotheses and attaches evidence that supports each one and evidence that argues against it. Every piece of evidence is labelled observed or inferred and tagged with its source: live metrics, demo target, seeded, or synthetic. The language model helps choose which graph tools to call, but hypotheses, evidence, and scores come from graph queries. If the evidence is too thin, Ripple says so and asks for more instead of guessing. Click any answer and you see the exact nodes, edges, and Cypher query behind it. Ripple remembers outcomes, not just events. It records whether each fix worked and under what conditions, and it checks every remembered fix against the current graph. A fix whose assumed dependency has since been removed is not recommended, and Ripple shows why. Risky actions need human approval, executions leave an audit record, and a verifier compares post-fix metrics with the pre-incident baseline over a time window. If data is missing, the outcome is inconclusive rather than a guess. Beyond incidents, Ripple answers “what would have broken if a different service had failed?”, with the uncertainty from unverified dependencies stated. A company-brain view answers who owns what and why decisions were made, with sources and warnings when a decision has been superseded. Each team gets its own isolated FalkorDB graph, and the tools are exposed over MCP so other agents can use the same graph. Ripple is for on-call, SRE, and platform engineers running microservices, and for the teams who inherit those systems. It covers all three tracks: agents that act on connected data, agent memory and coordination, and a company brain.

VibeSync·@xplorervishal

GridNudge - A Safety-Gated, Uplift-Aware EV Charging Decision System

Environmental Hacks/Oct 2026

VibeSync·@xplorervishal

PolicyLab: Prove Authorization Changes Before Production

First Commit/Sep 2026

https://builder.aws.com/content/3Jarm21NXl0AsHVg7PpHDLVTsM8/policylab-prove-authorization-changes-before-production

VibeSync·@xplorervishal