SolHive | Devpost

Inspiration

Solana was built for speed, but network congestion, failed transactions, and unpredictable fees often leave traders frustrated, missing out on golden opportunities. Imagine trying to buy an NFT at the perfect moment, only for the transaction to fail—or worse, paying sky-high fees due to congestion. We created SolHive to change that. By predicting congestion before it happens and analyzing NFT market trends in real time, our AI-powered platform ensures that users trade smarter, avoid costly mistakes, and seize the best opportunities—before everyone else. We set out to bridge this gap by combining AI-driven insights, predictive analytics, and real-time data aggregation, arming users with actionable information to make smarter decisions in the marketplace

What it does

The application provides real-time insights into Solana’s network performance by analyzing key metrics such as TPS, transaction fees, failed transactions, NFT trading volume, and minting activity. It leverages XGBoost with L2 regularization for time-series forecasting, predicting network congestion and suggesting optimal transaction times. A multi-agent AI system enhances decision-making by evaluating NFT trends through computer vision models (GPT-Vision, PyTorch), financial momentum analysis (SOL/USDT trends via The Graph), and sentiment analysis from aggregated news sources. The platform also scrapes and verifies Solana-related news using NLP models, ensuring users receive accurate, AI-driven market insights.

How we built it

We built the platform with a modular and scalable architecture, integrating real-time blockchain data retrieval, AI-driven analytics, and predictive modeling to provide actionable insights. On the frontend, we used React, JavaScript, and CSS to create a seamless and intuitive user experience. Privy authentication enables easy access through social media, email, or crypto wallets, lowering the barrier to entry for users. For data processing, we leveraged GraphQL, Dune, and The Graph to fetch key Solana network metrics, structuring the data for efficient analysis. The XGBoost model with L2 regularization was trained on historical blockchain data to predict network congestion and optimize transaction timing. The multi-agent AI system, built using reAct agents, powers NFT recommendations by combining computer vision models (GPT-Vision, PyTorch), price trend analysis (SOL/USDT), and sentiment analysis from web-scraped news sources. BeautifulSoup ensures accurate data aggregation, feeding verified insights into the AI engine. To support scalability and decentralized data retrieval, we integrated IPFS for NFT storage, ensuring reliability and accessibility across the ecosystem.

Challenges we ran into

Working under a tight deadline, one of our biggest challenges was optimizing real-time data retrieval while ensuring accuracy and efficiency. Fetching price data from The Graph required extensive documentation review to identify the most efficient Solana data endpoints. This process involved trial and error, but in the end, it was rewarding—allowing us to significantly improve latency and enhance the platform’s responsiveness. Balancing speed and reliability across multiple data sources was another hurdle, particularly when fine-tuning our XGBoost model to prevent overfitting within the limited timeframe. Despite these challenges, the experience deepened our understanding of blockchain data infrastructure and reinforced our ability to make rapid, informed optimizations.

Accomplishments that we're proud of

We’re proud of how we tackled complex challenges and delivered a high-performance platform within tight deadlines. Optimizing real-time data retrieval from The Graph significantly improved latency, enhancing responsiveness and efficiency. Successfully deploying an XGBoost-based congestion prediction model with L2 regularization ensured accurate forecasting while preventing overfitting. Additionally, building a multi-agent AI recommendation system that integrates computer vision, price momentum analysis, and sentiment detection was a major milestone, enabling intelligent NFT suggestions. Beyond the technical achievements, our ability to quickly adapt, integrate new technologies, and refine our approach under pressure was one of our biggest successes.

What we learned

This project was a deep learning experience, both technically and strategically. We gained a stronger understanding of real-time blockchain data retrieval, particularly optimizing The Graph endpoints to minimize latency. Building the XGBoost-based congestion prediction model reinforced best practices in time-series forecasting and regularization techniques to improve accuracy. Working with multi-agent AI systems for NFT recommendations taught us how to integrate computer vision, financial analysis, and sentiment detection into a cohesive decision-making engine. Beyond the technical skills, we learned the importance of rapid problem-solving, efficient documentation review, and agile development—critical for working within time constraints while maintaining high-quality outcomes.

What's next for SolHive

SolHive is planned to continue to evolve with especial focus enhancing our predictive model by incorporating deep learning techniques for even more accurate congestion forecasting. Our multi-agent AI system will be further optimized to provide personalized NFT recommendations, leveraging more diverse data sources. Additionally, we aim to integrate real-time alerts for network congestion and NFT market trends, ensuring users stay ahead of key developments. Expanding beyond Solana, we envision adapting our platform to support multi-chain analytics, making SolHive a comprehensive tool for blockchain intelligence.