Project Aegis (Anti-Gravity FinTech Edition)

Project Aegis is a simulated core-banking resilience and high-frequency continuity shield demonstration. It is built as a Multi-Agent System (MAS) via the Model Context Protocol (MCP) using NitroStack.

Important Disclaimer

This project is a simulation for demonstration purposes only. There is no real bank, no real money, and no real customer data anywhere in this system. It does not connect to any real banking infrastructure. Everything runs against a mock in-memory ledger initialized with synthetic data.

Related MCP server: MCP Business AI Transformation

Math vs. Presentation Flavor

To evaluate this project accurately, it is important to delineate the actual mathematical implementations from the presentation terminology:

Actual Mathematical & Systems Implementations:

• Incremental SVD Sketching: The IncrementalSVDEngine implements a real Frequent-Directions-style matrix sketch over a 4-dimensional telemetry vector. It operates in $O(k \cdot d)$ time, maintaining a baseline of nominal system behavior.

• Exponential Decay & Filtering: The sketch utilizes an exponential decay weighting ($\lambda = 0.95$) to adapt to volume growth and an $L_1$-norm clamp to reject extreme single-frame log corruptions.

• Anomaly Detection: The residualError(x) method mathematically computes the $L_2$ norm of the residual vector after projecting the current telemetry onto the healthy subspace: $|(I - P_S)x|$. Note: The anomaly threshold ($> 15.0$) is a tuned demo constant, not a derived statistical bound.

• Remediation Patterns:

  • Request Coalescing: The SingleFlightGate pattern safely deduplicates identical concurrent reads.

  • Idempotency Locking: The IdempotencyEnforcer calculates cryptographic hashes to mathematically reject duplicate submission attempts within a TTL window.

  • Admission Control: The QosShunting implements token-bucket based prioritization.

Presentation "Flavor":

• The terms "Anti-Gravity", "Orbital Drift", "Kinetic Mass", and "Structural Collapse" are presentation dressing to visually conceptualize standard load-shedding and telemetry-anomaly scenarios. They map directly to standard distributed systems concepts (e.g., "Orbital Drift" = SVD Subspace Residual Error, "Kinetic Mass" = Lock Contention).

• The agents (PRIME, ATLAS, CERBERUS, HERMES) act as standard MCP tool providers mapping to the remediation patterns above.

Demonstration Modes

Mock Mode (Default)

By default, the dashboard runs in "Mock mode", against an in-memory database generating synthetic, deterministic data.

Live Validation Mode

This project includes a real, verifiable proof layer alongside the mock engine. The Live validation mode tests the system against a genuine local Postgres 15 instance under real Artillery-generated HTTP load, proving that the anomaly detection is triggering on actual connection-pool contention, not scripted telemetry data.

To run the Live validation mode:

1. Ensure Docker and Artillery are installed locally: npm i -g artillery

2. Start the database: cd validation && docker-compose up -d

3. Start the legacy core server: node validation/legacy-core-server.js

4. On the frontend dashboard, toggle the "Simulations" data source to Live Validation Harness.

5. Triggering the Salary Day Storm will now spawn a child process executing the thundering-herd.yml load spec against the local Postgres server.

Note: While Live mode operates on real row-level ACID locks and connection pooling limits, it is still a local, single-machine demo rig and should not be considered a production core-banking environment.