TouchstoneIn development
AI coding agents are good at writing code but cannot tap through a real app on a real phone. Touchstone closes that gap: an agent orders a test over a REST API or MCP, human testers pick it up on real devices and return structured reports with video and screenshot evidence uploaded through presigned S3 URLs. The platform includes a full economy: budgets, multi-currency accounts with CBR FX conversion, a tester ledger, withdrawals and payouts via Alfa and Stripe. It also has a trust and safety layer with an evidence-analysis worker, report corroboration, tester streaks and honeypot tasks that catch farming.
Highlights
- — Exposed a pure REST API for agents (X-API-Key) and an MCP server with order_test/get_report, so agents order human tests like any other tool call
- — Modelled the task lifecycle as a state machine with slot matching, a review flow and versioned report schemas (report/v1 JSON)
- — Built the evidence pipeline: presigned S3 uploads of screen recordings and screenshots, analysed by a separate worker
- — Added trust and safety with report corroboration across testers, streak scoring and honeypot tasks that catch fraud
- — Built the full payment flow: top-ups through Alfa and Stripe, multi-currency accounts with CBR rates applied at approval, a ledger for testers, withdrawals and payouts
- — Backed Next.js dashboards with i18n using FastAPI, PostgreSQL and Alembic, a Redis worker and MinIO, deployed with Docker Compose and nginx on my VPS
Stack
Python · FastAPI · Next.js 14 · React 18 · PostgreSQL · Alembic · Redis · MinIO/S3 · MCP · Stripe · Docker · nginx