THESEFAWAY Products

Products

Software we build.

Alongside the security engineering, a small team and I build our own software. Some of it comes straight out of security work; some of it doesn’t. All of it is built the same way: private by default, honest about where it stands, and designed to be easy to use.

Product · 01macOS · Windows · CLI

Available

Trojan

A vulnerability scanner for codebases that runs entirely on your machine.

Trojan finds leaked secrets, risky dependencies and dangerous code patterns, then explains each finding in plain English with the exact fix. It adds AI-assisted penetration testing and one-click security and compliance reports for clients and investors, and it works with MCP.

  • Local-first: no source upload, works offline and air-gapped
  • Findings ranked into five levels by how exploitable and how damaging they are
  • Free scanner, with paid plans for AI testing and reports
Detection rules
3,400+
Updated weekly from public advisories
Languages
14
JS/TS, Python, Go, Ruby, Rust, Java and more
Median scan
4.2 s
On a 100,000-line repository
Code sent to servers
0 bytes
Scanning runs entirely on your machine

Product · 02Self-hosted · web app

Research prototype

PDFreak AI

Explainable PDF malware detection for small security teams.

Final-year research project, BSc Computer Science, Brunel University London, 2025.

Signature-based antivirus misses a lot of malicious PDFs, and most AI tools send your files to someone else’s cloud. PDFreak AI keeps everything self-hosted. Static and dynamic analysis extract 17 features, a Random Forest decides whether a file is malicious, and only flagged files go to a local Mistral 7B. That model uses retrieval over MITRE ATT&CK and a corpus of real malicious PDFs to name the threat, map the technique and recommend next steps in plain language.

  • Two-stage design: the language model only runs on files the classifier flags
  • No external API calls; containers built to move on-premises
  • Hardened intake: upload and decompression caps, timeouts, Zip Slip-safe extraction
PDFreak AI analysis pipelineAn uploaded PDF goes through static analysis (PDFiD, peepdf) and dynamic analysis (Ghidra, sandbox) in parallel. A Random Forest classifies the 17 extracted features. Benign files return a verdict; malicious files go to a self-hosted Mistral 7B with retrieval over MITRE ATT&CK and a malware corpus, which writes the report.maliciousbenignexplainsuploadPDF / ZIPPDFiD · peepdfstaticGhidra · sandboxdynamic17 featuresRandom ForestRAG · FAISS · ATT&CKMistral 7Bverdict · MITRE mapreportself-hosted models · no external API calls · classifier runs on CPU in under 2 s
Accuracy
83.2%
Random Forest on a held-out test set, ROC AUC 0.928
ATT&CK mapping
90%
Of 50 generated reports cited an appropriate technique
Classifier time
< 2 s
On CPU, no GPU needed
Training set
1,267
Benign and malicious PDFs, analysed in an isolated sandbox

In development

Morning

An AI morning companion for iPhone. It opens the moment your alarm ends and runs a short voice check-in: how you slept, anything on your mind, unfinished tasks. Then it helps you set one to three priorities, and the conversation saves itself as a journal entry.

iOS · voice-only sessions · on-device speech recognition

Available

Collana

One profile, real opportunities. Students, graduates and professionals build a single verified profile that recruiters search directly. Organisations describe the person they need in plain language and get a ranked shortlist back.

Web · Nova Scotia · collana.ca →

Get in touch.

Tell me what you run today and what you want to be true in three months. I will reply with whether I can help, and if not, who might.

Email
admin@thesefaway.com

There is no contact form on this site, so nothing you send is stored here. Email comes straight to me.