Contexer
Autonomous AI vibe-coding platform
Overview & Problem
Contexer is an autonomous development platform that moves beyond prompt-to-code iteration. It introduces a closed-loop agentic architecture where an autonomous agent (Viber) writes code, boots local runtimes, inspects DOM rendering via Playwright, catches runtime errors, and synthesizes code diffs for human approval.
The Engineering Problem
Traditional AI coding tools operate blindly: the developer prompts, copies code, manually runs the app, encounters an error, and pastes it back into the model. LLMs lack sensory observation of whether their generated code actually compiles, renders, or passes end-to-end user acceptance criteria.
What I Built
- ■Architected a context-driven engine where project requirements, user stories, and system constraints are established once to eliminate repetitive prompting.
- ■Developed Viber, an autonomous agent orchestrating prompt generation, multi-file code synthesis, and debugging pipelines.
- ■Integrated Playwright for automated sensory observation: capturing terminal logs, verifying DOM rendering, and running acceptance tests against user stories.
- ■Engineered an automated error resolution workflow that detects faults, generates targeted code diffs, and waits for user approval before applying changes.
- ■Structured the platform as a TypeScript monorepo across Next.js, Electron desktop runtime, Cloudflare Workers, and Cloudflare D1 SQL database.
Engineering & Architecture
Sensory Observation with Playwright
Connected the code synthesis engine to a headless browser test runner to validate UI rendering and application state before declaring a task complete.
Controlled Agentic State Cycles
Employed LangGraph to coordinate multi-agent transitions (planning → execution → observation → patch) with bounded recursion and human-in-the-loop gates.
Technology Stack
Results & Status
Actively under development as an end-to-end closed-loop autonomous software engineering environment.