Guardian Trader
Guardian Trader is a risk-first autonomous trading agent built to prioritize capital preservation above all else while still identifying sustainable profit opportunities. At its core is a deterministic Guardian layer that rigorously evaluates every trade before execution, enforcing strict risk controls, position limits, and predefined strategies to prevent unnecessary losses. The system emphasizes discipline, consistency, and transparency, ensuring that every decision is explainable and aligned with long-term objectives. By combining intelligent market analysis with rule-based validation, Guardian Trader adapts to changing conditions while maintaining stability. It simulates real-world trading performance with a strong focus on accountability, resilience, and continuous improvement over time.
Category tags:
.png&w=96&q=75){kind=small}
"1. Application of Technology: 4 / 5 Justification: Very solid, clean web development in TypeScript. The fallback mechanism (using hardcoded deterministic rules if the OPENAI_API_KEY is missing) is very smart for a hackathon demo. However, writing a high-frequency risk engine in TypeScript/Next.js routes rather than a persistent backend language like Python or Rust might cap its performance in a real high-throughput live environment. 2. Presentation: 4.5 / 5 Justification: Fantastic presentation. Being built entirely in Next.js means the UI and the application logic are tightly integrated natively. The concept of an AI agent generating "Trade Memos" that are then vetoed or approved by the Guardian layer makes the user interface very intuitive. 3. Business Value: 4.5 / 5 Justification: Very high. Emphasizing "capital preservation above all else" and explicitly branding as a system that uses "AI to assist, but never decide" is the smartest way to market an AI product to traditional finance and compliance officers. 4. Originality: 3.5 / 5 Justification: While beautifully executed, the underlying architecture is very similar to TreasuryShield (rigid rule limits blocking trade propositions) and the AI logic is a fairly standard gpt-4o-mini prompt to classify the market. It lacks the highly original Web3/Cryptography depth of STRIKER or the Walk-Forward Validation math of JudyAI. ⚖️ Pros & Cons Pros: Dual-Role Veto: The AI is strictly relegated to an "Analyst" role generating JSON market classifications, while execution is handled by deterministic static TypeScript thresholds. This guarantees the LLM cannot hallucinate a catastrophic YOLO trade. Excellent UX Sandbox: Built as a Next.js web application using a local Prisma Database makes this incredibly easy for users to simulate paper trading performance safely within their browser. Structured Analysis: Using gpt-4o-mini with a strictly enforced JSON-only prompt to require a thesis_summary and invalidation_summary before trading forces the AI to "show its work." Cons: TypeScript Constraints: Placing the core execution and backtesting loops inside Next.js components and API routes limits the complexity of the math that can be done compared to native Python implementations (pandas, numpy, scikit-learn). Basic AI Integration: The LLM implementation is relatively thin (a single fetch call to OpenAI) without the complex context-merging, multi-agent debates, or memory-retrieval graph logic present in higher-tier projects."
Sanem Avcil
For Innovators & Creators