Blocksight

The Problem: Most dApp and DAO operators are "flying blind." They see transactions on Etherscan, but they don't actually know who their users are. They can’t distinguish between a bot, a transient speculator, and a loyal power user. This lack of behavioral data makes retention and strategic growth nearly impossible in Web3.

The Solution: BlockSight is an AI-powered intelligence layer that transforms raw blockchain data into deep behavioral insights. We move beyond simple charts by providing an "Activity Index"—a time-decayed scoring system that measures the health and engagement of every wallet in your ecosystem.

Key Features:

Custom Indexing Engine: Users can submit any Ethereum contract address to trigger a deep-crawl of historical activity.

Behavioral Analytics: Advanced metrics including Gini Coefficients for wealth distribution, Herfindahl-Hirschman Index (HHI) for market concentration, and NFT diversity scores.

AI Intelligence Hub: A natural language interface powered by Gemini 3 Flash Preview that allows users to "chat with their contract." Ask questions like "Which whales are likely to churn?" or "What is the average gas spend of my most active 10%?" and get instant data-backed answers.

Spam Filtering: Automated detection of "airdrop spam" and illegitimate NFT collections to ensure analytics reflect real human activity.

BlockSight doesn't just show you what is happening on-chain; it tells you why it matters for your project's future.

The Stack: BlockSight is built with a high-performance modern stack: Next.js 16+ (App Router), Tailwind CSS 4, and TypeScript. For the data layer, we utilize Vercel Postgres with Drizzle ORM to handle complex relational schemas between users, indexing requests, and historical dashboard data.

The Intelligence Engine: The "magic" happens through our integration with the Vercel AI SDK and Google’s Gemini Pro. We implemented a sophisticated Tool Calling (Function Calling) architecture. When a user asks a question in the Chat Interface, the AI doesn't hallucinate; it dynamically calls our specialized internal analytics tools (getDappStats, getTopWallets, getTokenDistribution) to fetch real-time data from our Postgres database, summarizes it, and presents it in a human-readable format.

Data Science Logic: We didn't just use standard libraries. We built custom utility engines for:

The Activity Index: An algorithm using exponential decay (30-day half-life) to weight recent transactions more heavily than old ones, providing a "recency-aware" engagement score.

Economic Concentration: Real-time calculation of Gini coefficients and balance distribution buckets to identify protocol decentralization levels.

Hacky/Notable Details: To bypass the limitations of real-time multi-million transaction processing during a hackathon, we built a Vercel Cron-based queue system. This allows the platform to accept indexing requests, place them in a "PENDING" state, and process them asynchronously, mimicking a production-scale data pipeline while maintaining a snappy user experience. We also used Resend to implement an automated notification system that emails users the moment their custom data "X-Ray" is ready for viewing.