Public architecture notes from building MyInvestPilot — an AI-native investment system built through AI-driven development, constrained DSL design, hybrid agents, and a cost-aware solo cloud stack.
This repository is the external technical narrative of MyInvestPilot: design choices, trade-offs, and failures from building a production system where AI writes most implementation code.
It is not:
- a product user manual (that belongs to product docs)
- an internal AI blueprint repo (used for private planning/execution)
- a generic prompt cookbook
The goal is straightforward: document what it actually takes to operate as an AI-native system builder, not just ship demos.
From Rule-Based Scripts to AI-Native Engines: Why I Built a Constrained DSL
How constrained IR (JSON DSL), schema-driven contracts, and deterministic execution work together for reliable decision systems.
Why I Stopped Writing Code: The 60/40 Rule for AI-Native Engineering
How AI-driven solo development evolved from 60/40 alignment/execution toward 90/10 in many cycles, with human quality gates still mandatory.
The Agent Paradox: Why We Built a "Boring" Hybrid Architecture
Why we split Local Agent (flexible orchestration) and Remote Agent (deterministic async processing) instead of relying on a single autonomous agent.
Running a Multi-Vendor, Event-Driven Cloud Architecture as a Solo Builder
How a multi-vendor, async-first, serverless-friendly stack supports real usage with low operational cost.
A synthesized version of this series was published on the author's blog:
How I Built an AI-Native Quantitative System — bmpi.dev
This post distills the key architectural decisions across all four parts: constrained DSL design, schema supply chain, prompt evolution, and the hybrid agent paradox.
This repo is for builders working on:
- AI-native application architecture
- agent systems with reliability boundaries
- cloud-native/serverless systems for solo teams
I'm Dawei (@madawei2699), an independent builder. I previously shipped myGPTReader and invest-alchemy, which later evolved into MyInvestPilot.
MIT