Engineering

What we actually build.

One flagship product and three engineering capabilities. Everything below is our own work, described by what it does rather than by who bought it.

Start a Project Tell us about your project and its requirements.
01
Flagship product

HexaValt Cyber Range and Anti-AI CTF Engine

We’re developing a cyber range with 1,000 challenges, a 720-level Academy and a 120-level Campaign. The platform is not open to the public yet.

Cognitive honeypot traps

Artifacts carry plausible decoy tokens. Submitting one is recorded as a decoy hit instead of a solve, which exposes guesswork and automated scraping.

Rotating 90-second session nonces

Multi-stage challenges issue a nonce that expires after 90 seconds. Each stage must be answered inside the window, so a stale or replayed answer fails.

Kinetic canvas verification

The final step draws a one-time code on an animated canvas in the browser. It has to be read by a person at that moment and cannot be lifted from the page text.

Jailed multi-tool browser console

Each challenge opens a restricted /workspace console with tools such as nmap, strings, xxd, tshark and sqlite3. Directory changes and parent paths are blocked.

02
Capability

High-Availability Enterprise Network and Perimeter Architecture

Campus and enterprise network engineering built so that the loss of one provider, one firewall or one gateway does not take the site down.

Collapsed-core design

Core and distribution functions run on one redundant pair of switches, which keeps the topology small enough to reason about and to operate.

Dual-ISP BGP multihoming

Two upstream providers with BGP path control, so inbound and outbound traffic moves to the surviving link when one provider fails.

Active/standby ASA failover

Paired ASA firewalls in a failover cluster. The standby unit takes over the perimeter when the active unit or one of its links goes down.

HSRP gateway redundancy

Hosts point at a virtual gateway address shared by two devices, so losing one first-hop router does not strand a VLAN.

03
Capability

Network Intrusion Detection and Active Prevention (IDS/IPS)

Real-time L2-to-L7 packet inspection, anomaly detection, and automated inline mitigation engines built to detect and block active network-layer attacks before lateral movement occurs.

Layer-2 Spoofing and Rogue Host Defense

Real-time detection and active blocking of ARP cache poisoning, rogue DHCP starvation floods, and unauthorized DNS spoofing on local segments.

Volumetric and Slow-Rate DoS Mitigation

Threshold-based and behavioral state tracking to absorb TCP SYN floods, UDP amplification bursts, and low-and-slow HTTP connection exhaustion (Slowloris).

Custom Signature and Behavioral Rule Engine

Deep packet inspection (DPI) pipelines combining deterministic byte-pattern rules with protocol anomaly scoring to help operators prioritize alerts.

Automated Inline Quarantine and Telemetry

Instant firewall and interface rule injection to isolate hostile source MACs and IPs while streaming structured forensic packet logs for incident review.

Deep packet inspectionARP / DHCP / DNSDoS mitigationInline quarantine
Ask about IDS/IPS
04
Capability

Cryptanalysis, Binary and Packet Telemetry Tooling

Purpose-built Python tools for recovering keys, decoding unknown data, taking binaries apart and auditing network traffic in depth.

Cryptographic key recovery

Custom key-recovery engines, including a Meet-in-the-Middle attack on 2DES that trades memory for time to make a double-encryption key search practical.

Multi-cipher auto-decoders

Tools that try layered encodings and classical ciphers against unknown input and report the candidates that decode to sensible output.

Binary reverse engineering

Static and dynamic analysis of compiled programs to recover logic, embedded data and validation routines.

Deep network traffic auditing

Packet-level inspection of captures to reconstruct sessions, extract payloads and surface traffic that should not be there.

Case studies

How two builds came together

Only work we can show evidence for. Each study states what is finished, what is not, and where it falls short.

Case study 01

This website, built as a static, private-by-default release

Complete

Problem

A new company needed a credible website with no monthly hosting cost, no server to patch, and a way to review unreleased versions privately before anyone else sees them.

Objectives

  • No server-side code in the public request path
  • Strict security headers and no third-party scripts
  • Previews visible only after sign-in
  • Fast and accessible on an ordinary phone

Architecture

  1. VisitorBrowser over HTTPS
  2. Edge networkTLS, redirects, security headers
  3. Static releaseHTML, CSS and JavaScript modules
  1. ReviewerBrowser over HTTPS
  2. Sign-in gateOne-time code to an allowed address
  3. Preview buildMarked noindex, never in the sitemap
Python build stepHTML and CSSJavaScript modulesCloudflare PagesCloudflare Access
Implementation approach

The same page templates drive a local development server and a deterministic static build. The build rewrites wording for public release, adds content hashes for cache busting, allows the single structured-data block by its hash in the Content Security Policy, and refuses to write output if any template field is left unfilled. Tests build the site twice and require byte-identical output.

Verified outcome, limits and lessons

Outcome. No third-party JavaScript. A policy of default-src 'none' with explicit allowances. In local Lighthouse runs every page scores at least 95 for performance, accessibility, best practices and SEO.

Limits. Web fonts still load from Google Fonts, which receives visitors' IP addresses. The contact form is built and tested but not switched on yet, so email is the contact channel.

Lesson. Keeping the headers and redirects in the build, rather than in a dashboard, made them reviewable and testable like any other code.

  1. Shared design systemPalette copied from the CTF platform
  2. Static releasePrivate previews before launch
  3. Security PlaygroundFirst interactive module

Case study 02

A CTF platform where scores should mean something

In development

Problem

When an automated tool can solve a challenge from its text alone, a CTF score stops showing what a person can actually do.

Objectives

  • Different artifacts and flags for every learner
  • Guessed or scraped answers are recorded, not quietly accepted
  • Answers that expire, so replayed solutions fail
  • Certificates that anyone can check

Architecture

  1. LearnerBrowser and restricted console
  2. ApplicationFastAPI routes and sessions
  3. Challenge enginePer-learner seeding and checks
  4. StorageSQLite for progress and certificates
PythonFastAPISQLiteHMAC-SHA256
Implementation approach

Each learner receives seeded challenge files. Plausible decoy tokens are planted in artifacts and recorded as decoy hits when submitted. Multi-stage challenges issue a nonce that expires after 90 seconds. Certificates carry a serial number and a salted HMAC-SHA256 hash for verification.

Verified outcome, limits and lessons

Outcome. An automated suite of more than 1,000 tests covers the challenges, the Academy, the Campaign and certificates. The content covers 12 domains, a 720-level Academy and a 120-level Campaign.

Limits. The platform is not open to the public. How well it resists automated solving has not been independently measured or audited, and AI tools change quickly.

Lesson. Every answer path is an interface someone will try to automate. Checks tied to time and to what a person sees are harder to shortcut than static flags.

Have a build in mind?

Tell us what you’re building and what kind of help you need.

Start a Project