I used AI to pull together a retrospective from GitHub, and it turned out there was more progress than I’d imagined. Looking back on these two weeks, the point isn’t any single project but several threads moving forward at once.
MatrixTradingSystem
This stretch was mostly about pushing the trading system toward “something operators can genuinely trust.” A good number of features were added, and it’s now entering the wrap-up phase. It’s not just about features running — the flow has to be smoother, risk controls clearer, error messages understandable, and the risk of misclicks and misjudgments during operation reduced. Next up is shipping this one out.
AI Email Workbench
I also did a round on this alongside the AI PM course, pushing AI email operations from prototype toward something operable. AI isn’t just about generating drafts well; once it actually enters the workflow, you hit review, permissions, model switching, observability, SOPs, and safety boundaries — those are what really matter for landing it. It was fun to build, but it’s not there yet; it’s not something I can launch as my daily tool — the pain point probably isn’t obvious enough yet, so I’ll revisit it later.
Lectio
Lectio is live! This stretch felt more like moving from feature development toward an actual product: I added an update mechanism, bug reports, the BYOK experience, onboarding, permission flows, support/about pages, and a landing page. These aren’t the flashiest features, but if a product is going to be used by others, these things are a must. I genuinely find it a big help for reading documents and studying new tech — no need to switch between GPT and the document, or a web page (and if the page itself embeds AI like Gemini, the help is smaller). That said, the recent DuckDuckGo news was interesting to see — things tend to swing back the other way.
Mira (Personal AI Assistant)
Mira is still pretty early. This stretch was about multi-channel entry points, memory, tool orchestration, a judgment layer, and the infrastructure for an agent to safely observe, operate, replay, and hand off state. I increasingly feel the point of an agent isn’t just “can it get things done,” but what it did, how a human sees it, how to trace it, how to hand it off, and which actions need protection. For it to be truly useful, I need to grow its Skills and add Workflows — and that means I have to pin down the friction points in my own daily life and fill them in one by one, starting with things like calendar management and email management. Coding can wait; maybe I’ll build that after summer travel.
The smaller projects and cleanup kept moving too. The AIATCL practice site is mainly built for exam review; I’ve also signed up for the iPAS Associate and Intermediate AI Application Planner certification exams, and I plan to build another practice site for myself — aiming to earn both certificates this year. The chenfu.ai personal site organizes recent product work into a narrative others can understand, while the chenfu-kb knowledge base accumulates my own methodology, decision logic, and background knowledge that agents can use. There were also some device-side, voice-service, automated-data-maintenance, and security hardening tasks — they don’t look like the main battlefield, but they’re all about shoring up boundaries, stability, and long-term maintainability.
One more milestone: I finished AIA’s AI PM course! Two months went by fast.
To sum up, the overall feeling is that things are slowly turning from many separate projects into one whole system: real products, an operations console, agent infrastructure, a knowledge base, and an outward-facing presence. Plenty of it is still immature, but the direction keeps getting clearer — not just building AI demos, but putting AI / agent capabilities into real workflows, slowly growing them into product systems that can be used, maintained, and understood.
Comments & Feedback