AxiosChat: Your Intelligent Web3 Navigator

🎯 Inspiration

The inspiration for AxiosChat came from watching people we care about get left behind.

We've all seen it: a friend hears about crypto, gets excited, downloads a wallet -- and then just... stops. Not because they're not smart enough. Not because they don't want to learn. But because the moment they're faced with terms like "gas optimization" and "slippage tolerance," it feels like being handed a manual when all they wanted was to book a flight. Students who could use ChatGPT for homework but couldn't figure out how to swap tokens without watching five YouTube tutorials.

We witnessed countless individuals eager to explore Web3 but paralyzed by:

• That sinking feeling when you're three tutorials deep and still don't understand why you need five different apps just to trade tokens
• The paralyzing fear of clicking "confirm" on a transaction, knowing one wrong character in an address means your money vanishes forever
• The exhaustion of asking "simple" questions in Discord servers, only to get eye-rolls and "DYOR" responses
• The isolation of being told Web3 is "the future" while feeling like you need a computer science degree just to participate

The turning point came when we realized that conversational AI could be the bridge. If people can ask Siri about the weather or ChatGPT about recipes, why can't they ask an AI assistant to "send 10 USDC to my friend" or "explain what gas fees are"? AxiosChat was born from this vision: to democratize blockchain access by making it as natural as having a conversation.

💡 What It Does

AxiosChat transforms blockchain interaction and learning from a technical challenge into an intuitive conversation. It's your personal Web3 assistant that understands plain English and handles the complexity behind the scenes to give you what you want and remove all the other noise

Core Capabilities

Conversational Blockchain Operations
Users can execute real blockchain transactions using everyday language using Gemini 3.0. No need to understand hex addresses, gas limits, or ABI encodings. All of this was fine-tunned for the gemini model to easily understand the requirment so that it wont be too generic like the rest of other ai model. Simply type "send 10 ETH to 0x1234...." or "check my wallet balance," and AxiosChat handles the technical implementation.

Intelligent Context Understanding
The system doesn't just parse commands—it understands intent. We use Gemini ai to extract every intent of the user action where a user only wants to learn the plaform still provides that opportunity for them want to trade their token for them. I also added routes for individuales who just wants to learn or trade in the website so each process are deisnged for learners, technical or both sides to use it. Whether you ask "What's my ETH balance?", "How much Ethereum do I have?", or "Check my wallet," AxiosChat recognizes these as the same request and responds appropriately.

Real-Time Education
Every interaction becomes a learning opportunity. When you execute a function, AxiosChat explains what's happening, why gas fees exist, and what slippage means. Users gain knowledge organically while accomplishing their goals. Not just trading the ai teaches you what it is doing so if you start an interaction with someone youd be able to defend web3 actions in a way you understand through the frendley conveerstaion from gemini ai

Secure Transaction Management Critical operations requiring blockchain state changes are never executed automatically. Instead, they're queued for explicit user approval, displaying clear details about what will happen, estimated costs, and potential risks. Users maintain complete control. Since this are cash and we want to still allows the indivialdul to still have full control over every action taken by gemini we had to make sure that you as the user approves the action before gemini proced with it

Multi-Model AI Strategy
AxiosChat employs a sophisticated dual AI architecture:

• Primary conversational layer using the Gemini AI Network for blockchain specialized knowledge and interactions
  
• Specialized Web3 model for precise function calling and parameter extraction
  In order not to put everything into one AI that has to handle too many tasks at once and start losing context, we designed the architecture this way. We use a fine tuned Gemini model specifically for Web3 actions to avoid critical mistakes, and another Gemini model focused on friendly teaching and explanation for the user. Both models are then connected into a conversational system that combines their responses and

Example: I asked for a token, the mainstream identifies the purpose and sends it to the fine-tuned Gemini model for the required action, then sends it back to the conversational model. The fine-tuned model takes the data it processed and returns a response for the user while teaching them in the process.

Cross-Chain Vision
AxiosChat supports EVM networks out of the box. Built on a flexible, modular architecture, it seamlessly operates across Ethereum, Polygon, BNB Chain, Arbitrum, and other EVM-compatible networks, positioning AxiosChat as a truly universal Web3 interface.

🔨 How We Built It

Technical Architecture

Frontend Foundation We built the interface using React and TypeScript for type safety and component reusability. Tailwind CSS and Shadcn UI provide a modern, responsive design system that looks professional while remaining accessible. The UI features animated message bubbles, clear code formatting, and intuitive interaction patterns.

Web3 Integration Layer Thirdweb provides our Web3 connection layer, handling wallet connections, and transaction signing. We built a custom service layer on top that abstracts the complexity, allowing our fine-tuned gemini AI model to interact with blockchain functions

AI Architecture: The Intelligence Engine

The system operates through a carefully orchestrated flow:

Step 1: User Input Processing
When a user sends a message, it first reaches our conversational Gemini model. This model determines the user's intent and whether a blockchain operation is required.

Step 2: Intent Classification
The conversational model analyzes whether the request is informational ("What is DeFi?") or action-oriented ("Send tokens to this address"). Informational queries are sent to the models with prompts engineered to provide educational responses.

Step 3: Function Extraction
For action-oriented requests, a fine-tuned Gemini AI model extracts the exact blockchain function required and its parameters. It transforms "send 2 ETH to address" into a structured function call with the correct data.

Step 4: Action Type Determination
The system categorizes operations as either read-only (checking balances, viewing transactions) or state-changing (sending tokens). Read-only operations execute immediately. State-changing operations require user approval.

Step 5: Secure Execution
For state-changing operations, the data details are added to a queue where users can review, approve, or reject. Upon approval, the system executes the action securely through the connected wallet, providing real-time status updates.

Step 6: Result Presentation
Regardless of the operation type, results are sent to the conversational Gemini to provide appropriate context, explanations, and next steps in the chat interface.

Backend Integration
A Python backend service handles the fine-tuned function calling logic, processing requests and returning structured responses. The service communicates with the conversational Gemini AI through a clean API, maintaining separation of concerns and enabling independent scaling.

Development Challenges Overcome

Challenge 1: AI Hallucination Prevention
Early iterations sometimes generated invalid function calls or incorrect parameters. We solved this by implementing strict validation schemas and a two-stage AI process: one for understanding intent, another specifically trained for Web3 function extraction. We also individually split all the AI models and then connected them into one so that all of them could interact and perform the required actions they are designed for.

Challenge 2: Transaction Security
Balancing convenience with security was critical. We developed the transaction queue system after realizing auto-execution was too risky, even with AI verification. The queue provides transparency while maintaining conversational flow.

🚧 Challenges We Faced

Technical Hurdles

Blockchain Unpredictability
Gas prices fluctuate. During testing, some transactions failed and networks were congested—the blockchain's inherent unpredictability required building robust retry logic, clear error messaging, and user education about why certain operations might fail or take time, and I had to provide this information.

I was also faced with quota limits during testing, so I had to add rate limiting and splitting the AI helped reduce the quota issue. I started using multiple APIs to fix it.

Another challenge I faced was getting the learning materials for each part. What I did was provide the website learning materials: I used prompt engineering, scraped data from each website, and added it into the prompt design for each learning resource in the application.

🏆 Accomplishments We're Proud Of

Technical Excellence
Our dual-AI architecture demonstrates engineering. The intelligent coordination between conversational and fine-tuned models, combined with robust error handling and graceful degradation, showcases production-grade thinking.

Security Without Compromise
We maintained security best practices in the best way I understand to deliver a seamless user experience. The transaction queue system proves that security and convenience aren't mutually exclusive—they can enhance each other when designed thoughtfully. And I still believe more can be done for the security aspect of the project.

Real Problem Solving
AxiosChat addresses a genuine barrier to blockchain adoption. We're solving a fundamental problem that affects millions of potential Web3 users who currently feel excluded from this technology.

Educational Value
By integrating learning into every interaction, AxiosChat doesn't just enable transactions—it empowers users with knowledge. This creates a positive feedback loop where users become more confident and capable over time.

Extensible Architecture
We built with the future in mind. The modular design allows easy integration of new chains, protocols, and features without requiring architectural overhauls. This positions AxiosChat for long-term growth and evolution.

Flexible User Experience
We had to split it so that users who just want to use the app without the conversation can still use it, and those who are there for learning can still use it in its separate UI.

📚 What We Learned

Domain Knowledge Expansion

Web3 UX Is Still Nascent The blockchain space has focused heavily on technical capability while neglecting user experience. There's enormous opportunity for interfaces that prioritize accessibility and usability without compromising functionality.

Trust Requires Transparency Users need to see what's happening behind the scenes, especially with financial operations. Transparency builds trust more effectively than any amount of security certification or marketing.

Education and Execution Can Coexist We learned that teaching users while they accomplish tasks isn't just possible—it's preferable. Contextual learning creates better long-term user outcomes than separate educational resources.

🔮 What's Next for AxiosChat

Future Roadmap

Enhanced DeFi Integration
Deep integration with popular DeFi protocols—Uniswap, Aave, Compound, Curve—allowing users to provide liquidity, stake tokens, and earn yields through natural language. "Stake my ETH in the best yield farm" becomes a simple command.

Transaction History and Analytics
Implement conversational access to transaction history and portfolio analytics. Users can ask "Show me my biggest trades this month" or "What's my best-performing asset?" and receive clear, actionable insights.

NFT Management
Conversational NFT operations—minting, transferring, listing, and discovering. Users can manage their NFT portfolios through natural dialogue, with AxiosChat providing market context and recommendations.

Cross-Chain Abstraction Layer
Develop technology that allows seamless operations across non-EVM chains—Solana, Cosmos, Polkadot—presenting users with a unified Web3 interface regardless of underlying blockchain architecture.

Personalized AI Learning Tutor
Implement user-specific AI model fine-tuning based on transaction history and preferences. AxiosChat will learn individual user patterns and provide increasingly personalized learning responses.

Scaling and Infrastructure

Enterprise Partnerships Collaborate with wallets, exchanges, and protocols to integrate AxiosChat as their conversational interface layer, expanding reach and impact across the ecosystem.

---

Why Now

Addressing Real Pain Points
Unlike many blockchain projects solving niche technical problems, AxiosChat tackles the fundamental adoption barrier: accessibility. This is the problem that affects millions of potential users and limits blockchain's transformative potential.

Technical Sophistication
The dual-AI architecture, robust error handling, and thoughtful security implementation demonstrate engineering maturity beyond typical hackathon submissions. We've built something production-ready, not just a proof of concept.

Broad Impact Potential
AxiosChat's success would benefit the entire Web3 ecosystem. By lowering barriers to entry, we accelerate mainstream adoption, increase protocol usage, and bring new users who will explore beyond AxiosChat itself.

Educational Mission
By making Web3 education inseparable from Web3 usage, AxiosChat creates informed users who understand what they're doing and why. This elevates the entire space beyond speculative trading toward genuine utility.

---

Live Demonstration

Try AxiosChat: https://axioschat.vercel.app/

GitHub Repository: https://github.com/OkeyAmy/axioschat-sensay

For judges to try the project, you will need an Ethereum Sepolia Faucet:

You will have to switch to the Sepolia Testnet on the Function route, then copy your address and request tokens from the Google Cloud Web3 faucet: {https://cloud.google.com/application/web3/faucet/ethereum/sepolia}. Enter your address, then you will receive the tokens and can test it out without using mainnet tokens.

NB: You must be on the Sepolia Network for your faucet balance to reflect in the app.

---

Technical Details

Frontend Stack: React, TypeScript, Tailwind CSS, Shadcn UI, Framer Motion

Web3 Libraries: ThirdWeb

AI Integration: Google Gemini 3

Backend: Python with Flask, custom Gemini function calling service

UI Enhancements: React Syntax Highlighter, date-fns, custom error boundaries

Deployment: Vercel

---

Making blockchain technology as simple as having a conversation. This is Web3 without the gatekeeping. Without the fear. Without feeling like you're not "technical enough." Just you, your goals, and an assistant that actually helps.

Built With

• ai
• python
• thirdweb
• typescript
• web3