First Dollar

Overview

First Dollar is a platform where companies post bounties, content contests, and jobs, and creators and builders get paid for their work. Prize pools are escrowed in smart contracts on the Base chain, creators submit Instagram and TikTok content with verified ownership and live engagement tracking, and every wallet transaction is gasless for the user. It also includes Radar Room, a vetted product‑feedback community. Over 200,000 users signed up for the waitlist.

My Contributions

I worked across the full product — the web app and the smart contracts. The stack included Next.js + TypeScript for the frontend and APIs, Prisma + PostgreSQL for data, Privy for auth and embedded wallets, Solidity (Foundry) with Viem/Wagmi for the onchain layer, and Stripe for payments.

Bounty Smart Contracts: Built the escrow bounty contracts on Base — a factory that deploys a standalone escrow contract per bounty, holding the prize pool in ETH or any ERC‑20. The design separates roles: a Gnosis Safe multisig owns every admin and critical function (pause, refund, emergency withdraw, signer rotation), while a Privy server wallet authorizes finalizing winners and claiming rewards through off‑chain signatures — so no raw private key ever enters the app. All wallet transactions are gas‑sponsored, making the platform feel completely gasless to users.

Contract Safety: Every signature is time‑boxed with a 5‑minute expiry, replay‑protected by tracking used signatures on‑chain, and bound to the specific contract and function it authorizes. Payouts follow a pull‑payment pattern — winners are allocated claimable amounts and withdraw them themselves — instead of pushing funds. All fund‑moving paths are reentrancy‑guarded and pausable as a circuit breaker, deposits are measured by balance deltas so fee‑on‑transfer tokens can’t inflate the pool, and signer rotation is atomic with a guard against removing the last signer.

Payments: Implemented the platform‑fee model and migrated bounty funding to Stripe hosted checkout — publishing a bounty redirects the company to checkout, and a webhook marks it funded and moves it into review.

Social Verification & Engagement: Built two‑layer (client + server) ownership verification for Twitter, Instagram, and TikTok links — each link is resolved to its real author so spoofed submissions are caught. Also built engagement capture (views, likes, comments) across Instagram and TikTok, and video re‑hosting that powers a reels‑style submissions feed. Content‑contest submissions rank on a weighted engagement leaderboard — likes, retweets, comments, and quotes carry different weights, and engagement from unverified or brand‑new accounts is heavily discounted — which computes each creator’s proportional share of the prize pool.

Onchain Score: Built a 0–100 wallet‑reputation score computed from Base transaction history across six weighted dimensions — protocol diversity, activity consistency, log‑scaled volume, DeFi depth, wallet maturity, and builder signals — with multiplicative bot‑detection penalties for single‑contract farming, metronomic transaction timing, and wash trading. The score gates access to the Radar Room community and is surfaced in the UI with a per‑dimension breakdown and a GitHub‑style activity heatmap.

AI Feedback Scoring & Automated Payouts: Built the AI pipeline that turns raw Telegram feedback into payouts. Multi‑point messages are segmented into individually scorable pieces, classified across 11 labels (spam, AI‑generated, duplicate, actionable…), and scored on a strict four‑dimension rubric — relevance, actionability, specificity, and evidence — with vision models verifying attached screenshots and Gemini analyzing screen recordings. Embeddings cluster duplicate feedback by cosine similarity, penalizing copies and per‑user farming, and each tester’s top items aggregate with diminishing returns into a score that splits the project’s prize pool proportionally — no manual judging. Because the model’s output decides who gets paid, all user text is sanitized against prompt injection before it reaches the LLM.

AI Across the Product: AI verification that a bounty submission actually matches the brief (paired with GPTZero for AI‑text detection), auto‑generated feedback summaries and theme categorization for company dashboards, a chat assistant that interviews companies and emits a structured project brief, and AI‑formatted listings with automatic skill and category extraction.

Radar Room: Built the multi‑step application flow for the product‑feedback community — an onchain score gate, application questions, Telegram account linking with session‑based (MTProto) group management, manual admin approval, and email‑only invites.

Admin Tooling & Observability: Built a pending‑changes approval system where material edits to live listings are held for admin review with before/after diffs, plus an in‑house observability layer (logging, metrics, tracing) with Redis and Postgres sinks and an admin dashboard.