Personal Investment Income System

the problem

I had capital sitting idle and a hypothesis: could I generate reliable monthly income against it, deliberately, rather than opportunistically? The harder problem wasn't the strategy, it was me. My own trade history showed two investors living in one account: a research-driven one who wrote real theses before entering positions, and an impulsive one who showed up under boredom or a hot ticker and gave back what the first one earned. Any system I built had to manage the second investor, not just execute the first one's ideas.

the approach

I set a monthly income target as a percentage of total capital and built a local Flask app, backed by a structured Excel data store, to log every options and ETF trade with fields most trackers skip: thesis (required before entry), conviction score, an emotion flag, market context, exit plan, and a pre-committed max loss. If I couldn't fill in a one-sentence thesis, the rule was simple, don't take the trade.

On top of the log sits nine explicit, self-confirmed rules, things like never averaging down on a leveraged ETF, capping any single ticker at 15% of the portfolio, closing options at 80% of max profit instead of holding for the last scraps, and a rule stating the monthly income target itself is never a valid reason to force a bad trade. A Claude-powered assistant reviews the log against these rules, flags violations, and maintains a running "investor persona", a self-audit of which behaviors were actually generating income and which were noise dressed up as strategy.

what shipped

A dashboard tracking monthly income against target, live pricing via yfinance, and a rule-adherence view that surfaces violations before they compound rather than after a post-mortem. The persona document turned out to be the most useful artifact: reading back my own trade history through a consistent behavioral lens surfaced that a meaningful share of trades were self-admitted "rushed, no thesis" entries, and that averaging down on leveraged positions was my single highest-risk recurring pattern.

what I learned

The AI's value here wasn't picking better trades, it was enforcing pre-commitment. Writing a thesis and a max loss down before entering, instead of after the fact, is a small friction that changes behavior more than any model's "signal" would. Since the rules went into effect, income against the collateral pool has stayed positive every month, at or above target, and the system now catches rule breaks in near real time instead of surfacing them in hindsight.

The honest caveat: four months isn't a long enough track record to call this validated, and I'm treating it as an ongoing experiment, not a finished result. What I'd point to isn't the return, it's the pattern of building a governance layer around a decision process I already knew was inconsistent, which is the same problem enterprises hire for when they talk about "trust and guardrails" in AI deployments.