Skip to content
HG/os
aishipped

Aignt.fun

Aignt.fun — a production AI agent for DeFi strategy generation. I architected the complete strategy-generation pipeline on ExpressJS, AWS Bedrock, and Claude Haiku 4.5.

aiaws-bedrockprompt-engineeringdefi
Visit website →

Problem statement

DeFi users face a sprawling decision space — protocols, yields, risk profiles, and market conditions that shift constantly. Aignt.fun brings a conversational AI agent into the DApp to help users reason about and generate DeFi strategies.

Architecture

I architected the complete strategy-generation pipeline:

  • Model layer — AWS Bedrock serves Claude Haiku 4.5 as the agent's foundation model, chosen for low latency and cost at production scale.
  • Service layer — an ExpressJS service mediates between the DApp and Bedrock, handling conversation state, request shaping, and response validation.
  • Prompt engineering — structured prompt workflows steer the agent toward grounded, actionable DeFi strategy generation rather than open-ended chat.

Prompt strategy

The agent's behavior is governed by layered prompts: a system frame establishing the DeFi domain and guardrails, followed by task-specific templates for strategy generation. Conversational workflows were designed to keep responses grounded in the user's actual position and risk tolerance.

Production case study

This is the primary AI case study — a shipped product, not a prototype. Evaluation metrics, latency, cost optimization, and monitoring details are being expanded.