HyperValentPulse

HyperValentPulse

🔍 Description

HyperValentPulse transforms raw Hyperliquid blockchain transfers into actionable sentiment insights. Using AI, it classifies transactions as bullish, bearish or neutral - complete with reasoning. This helps traders, learners, and AI agents make sense of noisy on-chain activity in real time.

🧩 Hackathon Implementation

For the hackathon, we focused on building a V0 implementation that processes a 24-hour snapshot of Hyperliquid activity on 30th July 2025. All transactions from all blocks within that day were fetched and classified into bullish, bearish, or neutral, with reasoning provided by the AI. The MVP demonstrates how raw transaction data can be transformed into actionable sentiment insights, displayed in a minimal dashboard with token-based and sentiment classification based filters and aggregated statistics. This version validates the core use case - proving that big wallet movements and liquidity flows can be analyzed, classified, and surfaced for traders, learners, and AI agents in real time.

🛠️ How It Works

HyperValentPulse combines Covalent GoldRush foundational APIs with a locally hosted Llama 3.2 model to analyze Hyperliquid chain transactions.

1. Data Ingestion → Using Covalent’s Get Block Heights and Get All Transactions in a Block by Page (v3) GoldRush Foundational APIs, we fetch blocks, transactions, and logs (24 hour window i.e. 07/30/2025 - 07/31/2025).

2. Restructuring → Normalize raw JSON into a compact schema for sentiment analysis.

3. AI Classification → Each transaction is processed by Llama 3.2 (locally hosted it using Ollama) using a structured prompt, producing a JSON with classification, reasoning, and market impact.

4. Aggregation → Scripts generate summary statistics (counts, verdicts, token-level stats) for high-level dashboards.

5. Frontend Dashboard → A clean UI displays:

   • Verdict banner (Bullish/Bearish)

   • Charts (distribution & volumes)

   • Token/sentiment filters

   • Scrollable transaction feed with expandable reasoning

Technical Highlights

• Built on Covalent GoldRush APIs (Foundational + future Streaming).

• LLM prompt-engineered for structured sentiment classification.

• Produces two JSON outputs: llm_answers.json (per-tx) & statistics.json (aggregates).

• Designed for reuse by AI agents via structured APIs.

Technical Architecture & Data Flow

High-level flow:

Hyperliquid chain

   │

   ▼

Covalent  GoldRush foundational APIs

(Get Block Heights, Get All Transactions in a Block by Page v3)

   │  raw tx + logs (per block)

   ▼

Restructure/Normalize JSON

   │  minimal fields for analysis

   ▼

Local LLM (Llama 3.2) classification (Locally hosted and queried through Ollama)

   │  per-tx {classification, reasoning, market_impact}

   ▼

llm_answers.json  ───────►  statistics.json (aggregation script)

       │                               │

       └─────────────── UI (verdict, charts, filters, tx feed)

Covalent Foundational APIs at play:

1. Get Block Heights (day range → block numbers)

2. Get All Transactions in a Block by Page (v3) (paginate txs + logs)

 🔮 Future Outlook

While the MVP shows daily sentiment classification across all transactions, the vision is to make the system more focused, real-time and interactive:

• Historical Expansion – Extend beyond 24 hours to multi-day or monthly sentiment tracking (5, 10, 30 days). Users will be able to study how sentiment evolves over time.
• Real-Time Streaming – Integrate with Covalent GoldRush Streaming APIs and Hyperliquid infra for live dashboards that update continuously as transactions occur.
• “Whale Transactions” Filtering – Introduce a configurable filter where users define what counts as a “real” market-moving transaction (e.g., transfers above a set value threshold, whale wallet activity, liquidity adds/removals). This streamlines the signal-to-noise ratio by ignoring tiny dust transfers and surfacing only impactful moves.
• Cross-Referencing Historical Behavior – Combine the Streaming API with the Foundational API to analyze a wallet’s past activity. For example, distinguishing between a whale accumulation vs. a routine shuffle.
• Conversational Chatbot Interface – Embed an AI chatbot layer on top of the dataset. Users (traders, learners, analysts) could ask: “Why was this transfer marked bearish?” or “Show me bullish whale moves from the last 5 days.” The chatbot would leverage the structured LLM outputs to provide contextual answers.
• Autonomous Agent Integration – Expose structured JSON APIs that autonomous agents can consume for programmatic decision-making (trading, research, monitoring).

💻 Demo

👉  Hackathon Presentation: Google Drive 

👉 Additional UI Run Through (No Audio)

(Please reach out to pmtaday@gmail.com for any video related issues.)

📂 Project Links

• Code (GitHub): HyperValentPulse
  (Note (Highly Recommened: Navigate to the "ui" folder and follow the readme to locally host the UI).

• Docs/Notion (optional): Google Docs

👥 Team

•  Pushkar Mayur Taday (Telegram: @ptaday /GitHub: ptaday / Email- [email protected])