projects

Production agentic systems — built, shipped, governed.

Every system behind the practice. Own products, work built at ADF, and enablement for enterprise clients — described honestly, no invented metrics.

Flagship products05

eve.dev.lovinka.com

evelive

7 agents · survives SIGKILL · 38-tool HITL control plane

Architect·built at ADF

eve

A self-hosted agent platform that survives a hard kill — and resumes the human approval it was waiting on.

Seven agents — a root orchestrator plus migration, PR-review, Sentry-fix and outreach specialists — run on a Postgres-backed workflow world. A run paused mid-flight on a human-approval gate survives a hard SIGKILL and resumes on restart, with idempotent side effects. A live 38-tool control plane sets approval policy per tool; a React 19 observability console sits on top. No third party in the data path — LiteLLM routes to native Anthropic, native OpenAI and a self-made Codex-pool gateway.

React 19
Node 24
Postgres
LiteLLM
native Anthropic
Tauri / Rust

Read case study

fixit.app

FixItlive

Governed in-product AI · tri-stack RN / NestJS / Go

Founder · Architect·own product · LEFTEQ

FixIt

Real-time services marketplace with an in-product AI layer.

An Expo / React Native + NestJS + Go marketplace — recognise a problem from a photo, live tracking, in-app payments. Its in-product AI layer ships with real operational controls (budget, concurrency, idempotency, cost-tracking) and a Codex-pool gateway, on a multi-runner test pyramid (Playwright + Appium + unit/integration).

Expo / RN
NestJS
Go
PostGIS
Playwright
Appium

Read case study

deployik.lovinka.com

Deployik

~90 tools

→blue-green deploys
→auto-SSL · AES-256-GCM secrets
→agent-driven via MCP

Author·own product · LEFTEQ

Deployik

A self-hosted Go PaaS that agents can drive.

A single-binary, self-hosted platform-as-a-service in Go — blue-green deploys, auto-SSL, AES-256-GCM secrets — fronted by a TypeScript MCP server of ~90 tools so an agent can run the whole deploy / diagnose / screenshot loop in natural language. The site you're reading deploys on it.

Go
chi
go:embed
Docker
MCP (TS)
blue-green

Read case study

forge.app

Forgelive

Server-side agentic tool-loop → live bpmn-js canvas

Architect·an ADF product

Forge

AI-native BPMN authoring — an ADF product.

Forge — an ADF product — is a production AI BPMN-authoring capability: a from-scratch NestJS microservice running a server-side agentic tool-loop over SSE (function-calling + RAG) that drives a real bpmn-js Camunda-7 Designer canvas.

NestJS
bpmn-js
Camunda 7
OpenRouter
RAG
Angular

Read case study

reservine.io

Reservinelive

Own Angular UI library (73 components) · 19-provider AI backend

Founder · Frontend Architect·own product · co-founded

Reservine

Founder-built booking SaaS with a deep, multi-provider AI backend.

A multi-tenant reservation platform for any business. I founded it and own the Angular 21 Signals frontend and its 73-component UI library; the product runs 16+ AI services on a 19-provider LLM abstraction with structured output. Built with co-founder Martin Foltýn, who owns the Laravel backend.

Angular 21
own UI library
Nx
Laravel 12
multi-tenant
Stripe Connect

Read case study

Agentic systems & tooling03

lovinka.app

Lovinka

~20 agents

→kafka bus · per-service routing
→RAG context · confidence escalation
→human-approval trade gate

Author·own product · LEFTEQ

Lovinka

An autonomous multi-agent trading engine with a human-in-the-loop approval gate.

An autonomous crypto-trading engine: ~20 LLM-driven services on a Kafka bus, each with per-service model routing, confidence-based escalation, and RAG market context. Trade proposals queue behind a human-in-the-loop approval gate; a React control plane shows positions, the live AI-call log and the approval queue. Self-hosted end to end — TimescaleDB for the time series, no third party in the trade path.

NestJS
React 19
tRPC
Kafka
Python
TimescaleDB

Read case study

onyx.app

Onyx

use-without-reveal

→onyx:// refs → exec-time inject
→secret-scrubbed tool output
→per-item AI tiers · Touch-ID gate

Author·own product · in active build

Onyx

A zero-knowledge vault whose MCP server lets agents use secrets they never see.

A zero-knowledge secret vault with an MCP server built for agents: secrets are injected at execution time and never returned to the model (use-without-reveal), per-item AI tiers gate what an agent may touch, tool output is scrubbed of any leaked secret, and a kill-switch plus Touch-ID gate plus full audit log sit underneath. The 'safe agent control plane' made concrete.

Swift / SwiftUI
MCP
macOS Keychain
Touch ID
Node

Read case study

reservine-dx.app

Reservine-DX

GATE_RESULT: PASS

→11-phase delivery loop
→shell gate · non-zero on FAIL
→skills shared across repos

Author·own tooling · published

Reservine-DX

A published Claude Code orchestrator plugin whose delivery loop the model can't skip.

A published Claude Code marketplace plugin (@reservine/dx) of shared cross-repo skills — feature planning, cross-plan, implement-plan, cherry-pick, commit. Its 11-phase delivery loop is enforced by a real shell gate that emits GATE_RESULT: PASS/FAIL and exits non-zero on failure, so the agent literally cannot skip a phase. Agentic-engineering autonomy that's verifiable, not prose.

Claude Code plugin
skills
shell gate
TypeScript

Read case study

Client & enterprise01

AIX enablement

A major Czech bank

78.5% → 95.2%

→eval-harness quality delta
→refuse-and-escalate on risky changes
→DORA-aware PR review · ~10 repos

Consultant (via ADF)·external consultant — not on their dev team

A major Czech bank

Enabling a bank's developers to adopt agentic coding — safely.

As an external consultant via ADF, I built a major Czech bank's GitHub Copilot / AIX enablement kit — reusable skills, custom agents and prompts across ~10 repos, with banking-safety guardrails (auth / payment / key-management off-limits, anti-fabrication) and DORA-aware PR review. A reproducible eval harness — golden tasks mined from their own repos, A/B'd across models — proved the kit makes even weaker models refuse-and-escalate on risky changes instead of waving them through.

GitHub Copilot
AIX kit
eval harness
DORA-aware

Read case study