Why I Retired Three Projects to Focus Intensely on AI and Machine Learning.
Over the past two years, building across different domains was an invaluable part of my growth. I shipped DiagramPilot AI, OpenFolio AI, and an Automation Engine. Each solved genuine pain points and taught me system design.
Yet software is never truly finished. Every active project demands ongoing dependency audits, framework upgrades, and infrastructure upkeep. When attention is fractured across multiple distinct tools, maintenance rent silently consumes the hours required for mathematical depth and architectural rigor.
"When attention is divided across too many tools, maintenance rent silently consumes the hours required for genuine mathematical and architectural depth."
The Decisions: What Was Archived
Retiring working code is difficult, but pruning was necessary to move forward:
DiagramPilot AI: Great experience with Mermaid parsing and ASTs, but visual canvas utilities inevitably pull focus toward frontend styling rather than deep algorithmic systems.
OpenFolio AI: Building an interview-driven site builder taught serverless state management, but site generators quickly devolve into endless template tuning instead of data modeling.
Automation Engine: Connecting n8n workflows and webhooks helped daily productivity, but glue code does not advance fundamental machine learning knowledge.
The Singular Focus: Advanced Machine Learning & AI
Clearing away maintenance debt allows my technical energy to concentrate entirely on high-impact domains:
1. Graph ML & Network Topology: Modeling interconnected relational systems to uncover organized fraud rings that tabular models miss.
2. Causal Uplift Modeling: Moving beyond simple correlation to estimate individual treatment effects for targeted intervention.
3. Explainable AI (XAI): Implementing game-theoretic SHAP values to break down complex credit risk decision boundaries for high-stakes validation.
4. Time Series Forecasting: Engineering multi-step lag features and rolling statistical windows for multivariate demand prediction.
5. Production MLOps: Structuring reproducible model lifecycles with MLflow experiment tracking and Docker containers.
6. Deterministic AI Guardrails: Building concrete AST-level rules (AntiSlop & Luna) that stop autonomous coding agents from hallucinating UI slop.
Choosing Craft Over Clutter
A portfolio should reflect clear intent, not an attic containing every script ever written. Retiring these three projects was an exercise in deliberate focus: choosing depth over scattered breadth is the only way to build work that truly stands out.