Co-founder & CTO, Rinkt

Razvan Dimescu

I turn manual back-office operations into reliable software: RPA engines, workflow orchestration, and AI document pipelines, shipped for accounting, energy, and finance teams in Romania and the UK.

Currently Building Rinkt, an MCP-native RPA platform for back-offices. Licensed per automation, not per seat.

Discuss an automation → 20-minute intro call

Now · What I'm building

Day job: Rinkt, an MCP-native RPA platform for back-office work. Side project: Numa, a DNS resolver written from scratch in Rust that now runs as my system DNS.

Commercial · Paid work

Rinkt

MCP-native RPA platform for accounting firms and back-office teams.

Co-founder & CTO since 2021. Built the platform and every automation running on it: a cross-platform runner for browsers, desktops, and Office apps, with durable execution that recovers mid-step. Think UiPath, rebuilt so AI agents can call automations directly.

How we price it: licensed per deployed automation — you pay for the bots that run, not the seats that watch. A typical engagement starts at 3 automations per client.

Automations shipping on Rinkt

FacturaGPT

Invoice capture → eFactura / NIR posting, with mid-step recovery.

90% less manual entry facturagpt.ro ↗
Extras GPT

Bank statements across 9 RO banks → accounting codes.

80% less reconciliation extrasgpt.ro ↗
Online-payment allocation

Netopia & PayU payout files → matched receipts, posted nightly.

90% less manual handling custom · per-client

Go · FastAPI · Temporal · NATS · Playwright │ ● Active │ rinkt.com ↗

Open source · Side project

Numa

DNS you own. Everywhere you go.

Reach a service on your laptop from your phone or your team — without accounts, third parties, or giving up your keys. Portable resolver written from scratch in Rust — zero DNS libraries, wire protocol parsed by hand. Recursive resolution from root nameservers with full DNSSEC, ad blocking across 385K+ domains, DNS-over-HTTPS, .numa local domains with auto-TLS, LAN service discovery. My daily-driven system DNS.

Rust · tokio · axum · rustls │ ● Active · v0.13 │ numa.rs ↗ blog ↗ source ↗

Field Reports Case 01 — 04

Delivered via Rinkt 2021 → present

Case 01

Sector: E-commerce · UK distributor
Scope: Returns processing pipeline
Stack: OCR · GPT-4o · pgvector
Routing: Auto / Review / Escalate
Delivery: Via Rinkt

Unstructured email returns → confidence-based auto-processing.

Product returns arrive as inconsistent emails — PDFs, phone photos, varied formats. Built a pipeline that OCRs attachments, classifies issues with GPT-4o, matches products via vector similarity, and routes by confidence: auto-approve, review, or escalate.

Case 02

Sector: Finance · UK brokers
Scope: Quote aggregation · 10+ funders
Outcome: 30 min → seconds
Stack: Playwright · Go · NATS
Delivery: Via Rinkt

Vehicle finance rates across 10+ funders, in parallel.

UK finance brokers manually check rates across 10+ funders. Built parallel browser and API workers that query every funder concurrently and return unified quotes — replacing a 30-minute manual process with seconds.

Prior role Batiq SRL · AutoGrid · 2018 – 2025

Case 03

Sector: Energy · Grid services
Scope: Aggregation, dispatch, C&C
Duration: 7 years · POC → team of 8
Geography: France · Spain · EU grid
Role: Staff engineer · Batiq SRL

Solar aggregation, gas dispatch, frequency-containment reserves.

Started as a single-developer POC, grew into a team of 8 over seven years. Built aggregation software for solar farms totalling 5 GW, direct dispatch on negative energy prices with real-time notifications, and command & control for gas turbines producing 800 MW+. Deployed software and hardware for frequency-containment reserves across sites in France and Spain. Owned CI/CD, service deployment, SFTP configuration, and Helm charts throughout.

Case 04

Sector: Energy · RO distributor
Scope: Email triage · NLP
Training set: 36K messages · 212 categories
Accuracy: ~79% baseline
Role: Prior consultancy

Automated email routing across 212 internal categories.

A Romanian energy distributor routes thousands of customer and partner emails by hand. Built an ML pipeline (SVM + TF-IDF) that classifies incoming messages and extracts structured rules from their bodies — trained on 36K historical emails, now progressing toward production deployment.

Open source · Side projects 01 — 03

gopdf

PDF extraction & creation, pure Go.

Go · MIT │ ◌ Maintained │ source ↗

Parse, extract, and generate PDFs from Go without CGo or external libraries. Zero dependencies. Same protocol-layer instinct as Numa — read the spec, write the parser.

go-launcher

Auto-updates that don't brick themselves.

Go · MIT │ ◌ Maintained │ source ↗

Open-source Go library: external supervisor pattern, versioned directories, automatic rollback on failure. Think Squirrel, for Go desktop apps.

peekm

Markdown viewer with AI-session tracking.

Go · npm · Homebrew │ ◌ Maintained │ peekm.dev ↗

Single Go binary, installable via npm or Homebrew. Designed for reading long markdown + tracking AI-assisted coding sessions at a glance.

If your back office still runs on copy-paste and documents keyed in by hand, let's talk.

Discuss an automation →

razvan.dimescu@rinkt.com · LinkedIn · GitHub

Writing: Starting a logbook (Apr 2026) · archive at /writing/ · DNS pieces on numa.rs/blog.