What is This?

The Nexus Signal Engine is a real-world, agent-driven, adversarially resilient AI signal analysis and memory engine. It is designed for those who demand verifiable AI trust, audit-ready reasoning, and full adversarial resistance—not academic fluff or theoretical “AI safety.”

⸻

Key Features • Adversarial Obfuscation Defense: Recognizes and flags manipulated or obfuscated signals (tru/th, cha0s, leetspeak, etc.) using lemmatization, n-gram, and fuzzy-matching. • Agent Perspective Analysis: Every input is evaluated through multiple simulated ethical agents (Colleen, Luke, Kellyanne), each with its own reasoning lens. • Forensic, Immutable Memory: All records are cryptographically hashed, integrity-checked, timestamped, and stored in a rotating, full-text-searchable SQLite database. • Concurrency and Scale: Thread-safe with batch processing, auto-retry, database size limits, and lock management. Built to run reliably under real-world load. • Dynamic Configuration: Change risk thresholds, terms, and filters at runtime—no restart required. • Explainable, Auditable, and Tested: Full unit test suite included. Every signal is traceable, reproducible, and forensically sound.

⸻

Why Is This Different? • Not “just another AI filter.” This is the only open-source engine combining: • Real NLP (tokenization, lemmatization, n-gram defense) • Agent-lensed reasoning • Integrity-hashed, auto-pruned, fully auditable memory • Batch & concurrent processing • Production logging, auto-retry, config hot-reload • Tested for adversarial cases others ignore • Already published and timestamped on Zenodo: pip install -r requirements.txt python -m unittest discover

from nexissignalengine import NexisSignalEngine

engine = NexisSignalEngine(memory_path="signals.db") result = engine.process("tru/th hopee cha0s") # Obfuscated input print(result)

Requirements • Python 3.8+ • numpy • rapidfuzz • nltk • filelock • sqlite3 (standard lib) • tenacity

⸻

Citation

If you use or build on this work, cite the Zenodo archive and this GitHub repo:

@software{jonathanharrisonnexus_2025, author = {Jonathan Harrison}, title = {Nexus Signal Engine}, year = 2025, publisher = {Zenodo}, doi = {10.5281/zenodo.16269918}, url = {https://github.com/Raiff1982/Nexus-signal-engine} }

FAQ

Q: Why all the agent names? A: Each “agent” represents a different perspective in ethical or signal reasoning—making every decision more robust and auditable.

Q: Is this really adversarially robust? A: Yes—try to break it with obfuscation, leetspeak, prompt injection, or high-entropy word salad. Then check the test suite.

Q: Who built this? A: Jonathan Harrison (Raiff1982), published on Zenodo and open-sourced for the world to audit, use, or improve.

⸻

License

MIT License No hidden tricks. No “ethical” vaporware. Fork it, use it, cite it—just don’t pretend you built it first.

⸻

Contact

Questions, feedback, or press inquiries: Open a GitHub issue or reach out directly.

⸻

This project sets the bar for ethical AI signal integrity and memory. If you want to build trustworthy AI, start here—or play catch-up.