Stop guessing at mainnet — know why a transaction dropped, where the compute units went, and who decided the order of your slot.
You shipped an Anchor program and it works on devnet. Then mainnet: transactions silently drop, one hot account serialises everything you assumed was parallel, and the compute meter runs out mid-instruction — and the error message tells you none of it. This course goes below the program you write, into the machine that runs it. You will read the account model and Program Derived Addresses (PDAs) the way the runtime sees them, learn how Sealevel schedules non-conflicting transactions in parallel and what makes yours conflict, meter and price Compute Units so your transactions land during contention, study Jito bundles and Solana MEV defensively, write a Geyser plugin that streams account changes into your own indexer, put a million records behind one on-chain hash with state compression, ship confidential transfers with Token-2022, run a validator, and handle account-size caps, Cross-Program Invocation (CPI) depth, and upgrade authority once a program carries real value. The last module opens the agave source and adds a syscall to it. Five capstones — you pick one — finish with you changing Solana Virtual Machine (SVM) behaviour or running protocol-grade infrastructure on top of it.
Built by Lakshya Kumar
Paste this into any AI chat. Fill in the bracketed parts with your context — you'll get back a straight answer on whether this belongs on your plate.
I am learning Solana at the protocol layer — runtime architecture, Sealevel parallel execution, priority fees, JITO bundles (defensively), Geyser indexers, account compression, Token-2022 extensions, validator operations, custom programs at scale, and SVM internals. Help me understand the actual mechanics with reference to real agave source, real production protocols, and real SIMDs.
We grant free access case-by-case — students, career-switchers, builders on a tight budget. Sign in to send us a note.
Sign in to applyFinished the tasks? Take the prompt to your AI and get tested on it. We copy the prompt and open the app — just paste it in.
Declare an account writable and everyone touching it queues behind you — this is how you design programs that actually run in parallel.
Compute Units meter the work, priority fees buy the ordering, and getting either wrong means the leader drops you.
Solana has no public mempool, yet ordering is still for sale — learn the mechanics so you can build products that survive them.
Every Solana indexer starts at the validator’s plugin interface — write one and you stop paying for someone else’s stale data.
One 32-byte root can stand for a million records — the difference between an airdrop you can afford and one you cannot.
Balances nobody can read that the chain can still verify — plus the auditor key, and the production costs the docs gloss over.
What the machine actually does between gossip and voting — and why a validator that falls behind quietly stops getting credit.
Account-size caps, four-deep Cross-Program Invocation, upgrade authority, and the checks that keep an audit from finding your bug first.
Read the agave source, add a syscall of your own, and see why Eclipse, MagicBlock, and Firedancer all start from this same code.
Complete all modules, then submit the required number of capstone projects. Each must earn a passing rating from an admin reviewer.
Pick a cryptographic operation expensive in pure sBPF (Poseidon, Verkle primitive, Plonk verifier op). Write a SIMD draft following the SIMD template (motivation, specification, security considerations). Implement as both slow path (Solana program in pure sBPF) and fast path (fork of agave with new syscall registered). Benchmark CU cost. Submit the SIMD draft to solana-foundation/solana-improvement-documents for community feedback.
Write a Rust Geyser plugin (or Yellowstone-gRPC consumer) that watches Raydium, Orca, and Jupiter activity and streams price ticks + LP changes to Kafka/Postgres at <100ms latency. Handle reconnection, backpressure, and reorgs gracefully. Run for 24 hours and produce throughput + latency metrics. Bonus: include a small dashboard that visualizes the stream in real time.
Build a Bubblegum-style compressed merkle distributor with off-chain proof generation. Deploy to devnet with 1M entries. Implement claim flow with DAS-API proof fetching, replay protection (per-recipient bitmap), and a frontend that lets users check eligibility + claim. Benchmark cost-per-recipient and produce a one-page operations report comparing your design to uncompressed airdrops.
Build a Token-2022 program that issues a confidential SPL token, demonstrates encrypted-balance transfers among 10 recipients, includes an auditor-key reveal path (auditor can decrypt all transfers), and ships a working CLI/UI demo. Show on-chain inspection cannot reveal amounts but auditor can. README must explicitly state the trust model and limitations of confidential transfer in production.
Fork agave, modify block production for an ephemeral chain (MagicBlock-style or a session rollup). Demonstrate state settlement back to Solana mainnet via a bridge contract. Show a complete game/auction/trading flow where actions happen on the ephemeral rollup at <50ms latency and final state lands on mainnet at session end. Architecture doc + working code + 1-minute demo video.
Used throughout M6 for cNFT + compressed state design.