I finally am making my submission for the @SuperteamNG Advanced Infrastructure Challenge, powered by @SolinfraDev I worked on a project I named Sentry, a Solana transaction stack that uses Jito bundle submission with Yellowstone gRPC confirmations and an autonomous risk agent. For those who do not know, gRPC (gRPC Remote Procedure Call) is a modern, open-source framework initially created by Google. It allows an application to directly call a method on a server application on a different machine as if it were a local object. Inherently making fast and efficient for microservices and real-time systems. Sentry's backend is written in Rust, it builds bundles consisting of memo instructions and dynamic tip transfers. It simulates transactions locally before sending them through Jito "sendTransaction" endpoint to secure inclusion. My project tracks the lifecycle of each transaction from submitted through processed, confirmed and finalized, while collecting transaction timestamps at each step of the way. The confirmed and finalized delta comes in consistently around 8 seconds between runs, which gives a signal about validator voting depth under normal mainnet conditions. One major challenge in high-frequency transaction stacks is resource management. Running my project using Solinfra's infrastructure support, I had access to one concurrent gRPC stream, which means I could not simultaneously stream slots and verify transaction status. So I decided to offload slot updates to a background RPC loop, polling "getSlot" every 400ms, reserving the single Yellowstone gRPC stream exclusively to watch wallet transactions. To automate operational decisions using an AI agent, Sentry was built with an autonomous TypeScript agent. This agent runs a two-tier reasoning pipeline... It computes success rates and consensus delays from lifecycle log, then runs a local rules engine, then passes the structured context to a Groq LLM for a reasoned recommendation on whether to submit, hold, or retry. The entire pipeline can be controlled using an interactive CLI with a REPL and is monitored through a Next.js dashboard displaying live slot data and running telemetry. On a normal day, transaction stacks like what I built are individually common. What Sentry demonstrates is what could be possible when you build deliberate failover paths and keep transaction execution cleanly separated from AI reasoning. Practical applications could include arbitrage systems, liquidation engines, automated market makers and many more, where landing transactions in specific blocks is critical. A quick shoutout to @SuperteamNG and @SolinfraDev for this bounty. For developers building custom solana apps, you really should consider using SolInfra's Yellowstone gRPC combined with targeted RPC queries for managing stream limits and guaranteeing reliable confirmations under heavy network load. @solana @solana_devs #BuildInPublic Sentry's repo: github.com/SamuelOluwayomi/s… Sentry: sentryy.vercel.app Sentry doc: sentry-docs.vercel.app Solinfra: solinfra.dev Resources Jito TypeScript SDK github.com/jito-labs/jito-ts Jito Rust JSON-RPC SDK github.com/jito-labs/jito-ru… Jito Documentation docs.jito.wtf Yellowstone gRPC github.com/rpcpool/yellowsto… Triton Yellowstone Documentation docs.triton.one/project-yell… Solana JSON RPC API solana.com/docs/rpc Solana Documentation solana.com/docs