my crypto app stack
Backend dev, shipped my first crypto app (wallet reputation tool) over 4 months. Around 800 weekly active and $1.4k MRR right now. Full stack and resources below.
- Pick the chain first
Went Base because gas was cheap and Smart Wallet kills the "connect metamask" flow that scares non-crypto users. Solana commits you to its tooling early. Ethereum L1 only if users will pay $5+ gas without flinching.
- Stack
- Wallet: privy.io for embedded, wagmi + viem for read/write
- RPC: alchemy free tier to launch, paid around 50k req/day
- Indexer: ponder.sh self-hosted, skip Goldsky if cost-sensitive
- Frontend: nextjs 15
- Data layer
Dune is for analysis, not realtime reads. The Graph works but subgraphs hurt when you're new. Ponder listens to your contracts, writes to postgres, takes a week. Saves six.
- ML problem
At some point you'll need a prediction or classification model. Mine was sybil and scammer wallet detection before users interacted. Wrote my own heuristics for a month, they were bad.
Registered my app on pond and sent a bounty for help. Their Model Factory lets you put up the spec and a payout pool, the community submits trained models in 48h, you pick whichever scores best on your eval set and pay them out.
Got 30+ submissions, top model hit 0.91 precision, paid out roughly $400 total. Made it way cheaper than an ML contractor and didn't need to learn graph nets myself.
- Resources
- https://joinpond.ai/bounties - pond
- buildspace older course archives
- viem docs over ethers (cleaner)
- https://learnweb3.io/for concepts
- https://speedrunethereum.com/
- Skip
- Hardhat, use Foundry
- Truffle
- Writing your own indexer with getLogs()
- Reading OpenZeppelin docs cover to cover
TLDR: cheap chain, paid indexer, embedded wallets, post a bounty for any ML you need instead of training it yourself.
happy to answer specifics.