Frame
Define the user, system boundary, data, constraints, and measurable acceptance.
Proworks AI Systems · Bucharest / Remote
Production AI systems, built beyond the demo.
MCP servers, agent backends, RAG systems, and workflow integrations—with tests, approvals, failure handling, monitoring, and documentation.
Fixed entry engagement€1,500 · 5 business days
SYSTEM / LIVEPRODUCTION BOUNDARY
01 / INPUTRequest + contextvalidated · scoped
02 / CONTROLAgent runtimebudget · approval · state
03 / TOOLSAPIs + dataleast privilege · logged
01 / Services
One boundary. One accountable build.
Choose the smallest engagement that resolves the current technical uncertainty. Every implementation includes testing, deployment, documentation, and operational handoff.
01
Architecture Review
A paid decision sprint before anyone commits to a build.
€1,500 fixed↗
02AI Integration
Connect models to the systems where work actually happens.
€6,000–€12,000↗
03Custom MCP Server
Expose real tools and data to AI clients through a controlled interface.
€6,000–€12,000↗
04Agent Systems
Agents with budgets, approvals, auditability, and a way to stop.
€12,000–€35,000↗
05RAG Systems
Private knowledge retrieval measured before it reaches users.
€12,000–€30,000↗
06Production Hardening
Make the prototype safe, observable, and supportable.
€3,000–€8,000↗
02 / Proof
Inspect systems, not adjectives.
Public source and concrete architecture are more useful than a wall of logos. These projects demonstrate network safety, real-time agent control, and integrated product delivery.
MCP · Network safetyRE
CASE / 01
readable-mcp
A production-oriented MCP server that turns permitted web pages into clean Markdown while treating network access as a security boundary.
Open system →Agent systems · MobileTE
CASE / 02
Tether
A native Android client and real-time backend for operating Claude Code remotely over a private Tailscale network.
Open system →AI product · Content operationsA-
CASE / 03
A-Stats
One operating surface connecting keyword research, structured drafting, publishing, distribution, and search feedback.
Open system →03 / Delivery
Production is a set of controls.
A model call is the easy part. The engineering work is in permissions, boundaries, evaluation, recoverability, cost, and handoff.
Design
Choose the simplest architecture and make failure behavior explicit.
Build
Ship in reviewable slices with tests, evaluation, and observability.
Operate
Deploy, document, train, and leave a system your team can own.
04 / Open source
All repositoriesReusable patterns, in public.
Pythonreadable-mcp
Safe URL-to-Markdown extraction for AI clients.
↗Pythonmcp-job-queueDurable background work for long-running MCP tools.
↗TypeScriptagent-orchestratorExplicit coordination patterns for bounded agent work.
↗TypeScriptssrf-safe-fetchReusable network safety controls for server-side fetching.
↗Paid entry engagement · €1,500
Start with the system boundary.
The Architecture Review turns a use case, prototype, or delivery opportunity into a defensible implementation plan in five business days.