How I actually build these systems — architecture decisions, costs, failures, and results. No theory, no hype. If a post is marked "soon", it's being written between client work.
How one person takes a B2B marketplace from empty repo to live platform with real companies and press coverage: scoping the frozen fruit trade, building marketplace mechanics (listings, matching, messaging), onboarding an industry that lives on the phone. Full write-up coming.
A content pipeline that lives entirely inside Telegram: research → structure → draft → format → publish, on schedule, with zero clicks. Architecture, prompt design, failure handling, and per-post running costs. Full write-up coming.
A network of coordinated agents running a complete SEO operation — keyword research, briefs, drafting, internal linking, publishing, rank monitoring. Being built in public: devlogs will land here as it progresses.
The full pipeline behind my blog bots — architecture, prompt design, failure handling, and what it costs to run per post.
Why one agent isn't enough, how the agents divide the work, and where humans stay in the loop.
Define the eval set before the pipeline, then make the numbers go up. The process I use on client RAG systems.
Lessons from AgroB2B Hub that transfer directly to 0→1 AI work: scope ruthlessly, launch embarrassingly early, talk to users daily.