Crypto Copilot

AI trading demos often collapse into two bad patterns: vague chatbot advice or unsafe agent places trades automation. Crypto Copilot explores a better architecture: use AI for structured decision support while keeping execution constrained, inspectable, and simulation-only. The engineering challenge is to connect market context, typed APIs, schema-constrained LLM output, validation guardrails, and a frontend workflow without pretending the prototype is a production trading platform.

• Typed FastAPI API boundary for market data, LLM decisions, and simulated execution.
• Pydantic decision contracts for proposed trades, cancels, stop moves, and flatten actions.
• Market data normalization for candles, order books, and recent trades.
• BYBIT spot market context integration through CCXT.
• Read-only DEX pool snapshot endpoints for Uniswap v3 and Meteora.
• Protected showcase endpoints using X-Demo-Api-Key.
• Async SQLAlchemy and Alembic provide migration-backed persistence scaffolding.
• Separate Next.js platform repository demonstrates dashboard and plan-review surfaces without turning the project into a single backend script.
• Simulation-only execution model; no live trading, broker writes, or exchange writes.

The AI component is treated as a decision-support module, not an autonomous trader. It receives structured market context and produces a schema-constrained trade plan. That output is validated before it can move into the execution simulator. This is the correct boundary for a credible financial AI prototype: AI can reason over context, the API constrains output shape, guardrails validate the plan, execution remains simulated, and no real credentials or exchange-write paths are required.

Crypto Copilot intentionally does not implement live trading, account management, broker writes, production-grade user authentication, or production trading guarantees. That restraint is part of the architecture. The project focuses on backend/API judgment, safe AI integration, typed contracts, market-data normalization, and trading workflow design while keeping real-money execution outside the system.

Crypto Copilot demonstrates that I can build AI-assisted systems in a domain where correctness and safety matter. It connects my professional exposure to crypto and trading environments with product engineering: market data ingestion, typed service design, structured AI output, validation, simulation boundaries, and a user-facing copilot workflow.

• Add a portfolio-facing architecture diagram.
• Add persisted AI decision audit trails.
• Add richer simulation reports.
• Add exchange adapter abstraction.
• Add stronger evaluation fixtures for AI plan quality.
• Add frontend flows around plan review and risk explanation.

• The API README defines the simulation-first showcase flow and explicitly states no live trading, no exchange writes, no production auth, and no full production readiness.
• pyproject.toml documents FastAPI, Pydantic, SQLAlchemy, Alembic, CCXT, OpenAI, Web3, HTTPX, WebSockets, Pytest, Ruff, Black, and mypy.
• app/main.py wires health, candles, market data, DEX, LLM decision, and execution-simulation route groups.
• app/schemas/llm_contract.py and app/api/routes/llm_decider.py define the strict decision contract and protected LLM decision route.
• app/api/routes/exec_sim.py keeps execution in protected simulation, while DEX routes expose read-only pool snapshots.
• The platform package.json and app/services files support the Next.js interface shell, React Query client calls, dashboard status, positions, orders, PnL, and prompt surfaces.