AIGIS release candidate · accepting pilot partners

A valid signature is not an authorization.

AIGIS is an isolated verification layer between your backend and your payment gateway. Every outgoing payout is checked independently before money moves. Anything that cannot be confirmed goes to manual review instead of out the door.

See the shadow-mode pilot How it works

Built by Uravin d.o.o. (Montenegro). Our team has delivered software, business & systems analysis and consulting since 2019.

The problem

The biggest recent losses did not need stolen keys.

Attackers compromise one internal service or one set of partner credentials and send the payment system commands that look perfectly legitimate. The system signs and executes them, because that is what it was told to trust.

Crypto exchange · Sep 2026

≈ $387M

Bitget. Reported: a compromised backend in the wallet infrastructure spoofed transaction data, and the exchange’s own authorization process approved the withdrawals. Private keys were not compromised.

Real-time payments · Jul 2025

R$800M+

Brazil, Pix infrastructure. Reported: valid client credentials at a technology provider were used to create and sign fraudulent transfer orders out of several institutions. The central bank tightened rules for the whole sector.

Lenders & fintechs · Oct 2026

~20 institutions

India, CERT-In advisory. Reported: stolen payment-aggregator and partner-bank API credentials pushed payouts past KYC, loan-sanction and authorization checks. Recommended: maker-checker, limits, reconciliation against approved records.

The question is no longer “is this command signed?” but “should this money move?”

Read the full breakdown of these incidents and a checklist for your payout flow →

AIGIS

Independent verification for every outgoing payout.

AIGIS sits outside your main backend, with its own keys, rules and state. A compromised backend can ask for a payout, but it can no longer approve one on its own.

Authentic commands only

Each payout command is signed and checked against a registry of trusted keys. A changed amount, a changed recipient, an unknown key or a missing signature is rejected. Signatures use post-quantum ML-DSA (Dilithium).

Signature ≠ authorization

A correctly signed command still has to pass your policy: limits, beneficiary rules and a match against the approved record it claims to pay. Anything unconfirmed becomes REVIEW_REQUIRED.

No replay, no rollback

Ledger versions only move forward. A repeated, skipped or rolled-back version is rejected, so an intercepted request cannot simply be sent twice.

Tamper-evident evidence

Every decision is written to an append-only log linked by a hash chain, with the evidence behind it. History cannot be rewritten quietly.

Formal checks before execution

Business rules are expressed as invariants that the architecture checks with the Z3 solver before a transfer is released, instead of searching for holes after the fact.

Fail-closed by default

On any doubt, a lost connection or an unknown outcome, AIGIS holds the payment and asks for reconciliation of the original operation rather than retrying it.

Who it is for

Payment service providers and payout platforms
E-money institutions and digital wallets
Digital lenders with automated disbursement
Licensed iGaming operators paying out winnings
Real-time payment connectivity providers
Digital-asset custodians and exchanges

Where it stands today

Core written in Rust, covered by level-by-level security tests: forged signatures, tampered amounts, recipients and balances, replay, rollback and lost-connection scenarios.
Release candidate. We are onboarding a small number of pilot partners.
Connection to your balance source and settlement is built per partner during the pilot.

We will show you the test run live on a call.

Shadow-mode pilot

Two to four weeks. Nothing is blocked.

You see what AIGIS would have stopped on your real payout flow before you let it stop anything.

Scope

A 30-minute call to pick one payout flow and agree which rules matter: limits, beneficiaries, approved records.

Deploy in your perimeter

AIGIS runs inside your infrastructure and reads a copy of the payout stream. Your data does not leave your network.

Observe

Every command is verified in parallel with production. Nothing is held, delayed or changed.

Report

Which payouts would have been held and why, plus where your current controls can be bypassed. You decide what comes next.

Discuss a pilot

Services

Engineering and analysis since 2019.

AIGIS grew out of client work. Our team has built and analysed systems for banks, retail chains, pharmacy networks and city infrastructure since 2019, and we still take on that work.

Software development

From discovery to production with a small senior team.

Integrations: REST/SOAP, Kafka, RabbitMQ, EDI, e-invoicing, ERP
Data pipelines and PostgreSQL
AI and agentic automation with LLM quality gates

Business & systems analysis

We go to the raw data first and turn findings into requirements engineers can ship.

AS-IS / TO-BE process models in BPMN
Decision rules in DMN
Specifications, acceptance criteria, UAT

Consulting

Architecture and delivery for regulated environments.

Solution and integration architecture
BPM platforms on Camunda 8, zero-downtime operations
Compliance: Track & Trace, GDPR, EU AI Act

Selected work

Projects

Banking · Montenegro2025–2026

CKB Banka (OTP Group): BPM platform on Camunda 8

Embedded with the bank’s compliance analysts, DevOps and developers. We turned compliance and business requirements directly into executable BPMN processes and DMN decision tables, took KYC and customer onboarding, lending and compliance checks to production, and owned the platform roadmap.

Ran the platform in a live banking environment with zero downtime, including a production cluster migration from Camunda 8.6 to 8.8 with full rollback readiness.

Camunda 8 / ZeebeBPMN / DMNKubernetesKibanaPrometheus

Zerodowntime in live production

8.6 → 8.8live cluster migration

KYConboarding, lending, compliance

CIOwritten reference from the bank

Retail · supply chain2026

Procurement visibility for a multi-entity retail chain

738 SKUs flagged on day one

Unified eight messy sources (ERP via OData, e-invoices, supplier price lists, order chats) into one model of orders, deliveries and stock. Orders are calculated from real sales velocity including lost demand. Every line is tracked from order to invoice to delivery.

On day one it found that missing stock was not a store problem: a supplier was silently omitting ordered items. Read-only towards the ERP, on-premise, a new store connected in one day.

PythonPostgreSQLODataE-invoicing

Pharma & grocery retail · compliance2019–2022

Track & Trace serialization for national retail chains

MVP in 6 weeks for 1,500 pharmacies

Regulatory compliance MVP for a national pharmacy chain, delivered against a hard legal deadline: requirements with lawyers and IT, architecture, vendor coordination with SAP and the regulator API, phased rollout.

Rollout architecture for a grocery retailer with 20,000 stores, managing a multi-vendor programme across SAP, regulator and retail IT, with a team across Germany and India.

SAP ATTPS/4HANAGS1 / EPCISKafkaEDI

City infrastructure2022–2024

Decorative lighting management for a megacity

300,000+ managed assets

Delivery lead and solution architect for the system that runs a megacity’s decorative lighting across five municipal departments. AS-IS / TO-BE process architecture in BPMN, 18 automated workflows, integration contracts and product metrics.

Moved more than 50 users from paper and messenger chats to tracked digital workflows.

BPMNWorkflow automationIntegration architecture

AdTech · computer visionR&D

Real-time audience recognition for contextual advertising

Live gender & age detection

A computer-vision system that recognised the audience’s gender and age in real time and selected the advertisement to match.

Computer visionReal-time processing

Own product · B2B2025–present

mandarin.business: supplier intelligence platform

Live in production

B2B platform for finding and verifying suppliers in China: catalog ingestion and normalisation, supplier risk checks, automated research pipelines with multi-provider LLM routing, GDPR-compliant architecture with an on-premise option.

Next.jsn8nOpenRouterPostgreSQL

mandarin.business →

Own product · AI2025–present

Cainix: AI pipeline for manuscripts and screenplays

LLM quality gate with audit trail

Multi-step agentic pipeline for manuscript analysis and screenplay generation. Every output passes an evaluation layer with section-level approval, evidence requirements and manual override before it reaches the user — the same “verify before release” idea behind AIGIS.

TypeScriptClaudeOpenAIn8n

cainix.com →

About

Uravin d.o.o.

A software engineering and consulting company based in Montenegro. We work with banks, fintechs and retailers that need systems which behave correctly under pressure: in production, under audit and under attack.

Behind every project is a team of senior engineers, systems and business analysts, and DevOps specialists who have shipped software for banks, national retail chains and government-scale systems. We assemble the right people for each engagement, and the same people stay with it from discovery to production.

CompanyUravin d.o.o.
RegisteredMontenegro, 2022
Team experienceSince 2019
TeamEngineers, analysts, DevOps
ContractsEU-compatible B2B, EUR invoicing
Working hoursCET, remote and on-site
LanguagesEnglish · Russian · Montenegrin
Flagship productAIGIS

Leadership

Who leads what

Sergei Vasiliev, founder and CEO of Uravin

Sergei Vasiliev
Founder & CEO

Leadership · architecture · delivery

Twenty years in B2B software: product director at a national fiscal data operator (five B2B SaaS products, 120,000+ business clients), enterprise Track & Trace delivery, a banking BPM platform on Camunda 8 and, most recently, AI and agentic systems in production. Physics, Lomonosov Moscow State University.

Anton Vylderov
Head of Engineering

AI systems · security · AIGIS core

AI systems engineer and architect of secure AI platforms: multi-agent systems, post-quantum cryptography and transaction protection in AIGIS, medical decision-support AI and AI-driven CAD/CAM. Security background in OSINT, red teaming and SecOps. Python, Rust, FastAPI, PostgreSQL, Docker; OpenAI, Claude, Gemini / Vertex AI, LangChain, RAG, computer vision.

Eduard Shikalov
Head of Growth

Marketing · acquisition · AI automation

AI architect focused on business process automation and growth: digital marketing, SEO, paid acquisition, website development and rollout of AI tools. Master’s degree in Accounting and Finance, specialising in Banking, Donetsk National University of Economics and Trade.

FAQ

Questions we hear first

What is AIGIS?

AIGIS is an independent verification layer that sits between your backend and your payment gateway. It checks every outgoing payout command against its own keys, rules and state before money moves, and holds anything it cannot confirm for manual review.

Why is a valid signature not enough to authorize a payout?

In recent incidents attackers did not steal signing keys. They compromised a backend service or partner credentials and sent commands that the system signed and executed as legitimate. A signature proves who sent a message, not that the money should move. AIGIS checks both separately.

Is AIGIS a payment or settlement system?

No. AIGIS does not execute payments, hold funds or replace your core banking system, HSM or anti-fraud tools. It is a control layer: it decides whether a payout command may proceed, and only your bank or payment provider confirms balances and final settlement.

What happens if AIGIS cannot verify a payout?

The payout is held in REVIEW_REQUIRED. On any doubt, lost connection or unknown outcome AIGIS fails closed: it does not retry or resend the payment, it asks for reconciliation of the original operation.

Which post-quantum algorithms does AIGIS use?

Payout commands are signed and verified with ML-DSA (Dilithium), standardised by NIST in FIPS 204. Hybrid post-quantum TLS with ML-KEM (FIPS 203) protects the channel.

How does a pilot work?

A shadow-mode pilot runs inside your infrastructure on a copy of one payout stream. Nothing is blocked or delayed. At the end you get a report of which payouts AIGIS would have held and why. Your data does not leave your network.

Is AIGIS production-ready or certified?

AIGIS is a release candidate tested with automated and adversarial test suites in isolated environments. It is not certified by any regulator, and connection to a partner's balance source and settlement is built during the pilot.

Contact

Let’s look at your payout flow.

Tell us how money leaves your system today. We will tell you honestly whether AIGIS fits, and what a shadow-mode pilot would show.

info@uravin.me Telegram

+382 69 741 552 LinkedIn