Agentic AI · RAG Systems · Risk & Decision Intelligence
— "The interesting part of an AI system is what happens when it's wrong."
| 32 Public Repositories |
Python & C++ Core Languages |
RAG + Agents Core Discipline |
Tallahassee, FL Based In |
I'm a Senior AI & Machine Learning Engineer working across agentic systems, retrieval-augmented generation, and risk/decision models — the kind of software that has to keep running correctly after the demo is over.
Most of my repositories share a common discipline: define the failure modes first, then build the system that survives them.
flowchart TD
A["🎯 Problem & Invariant Definition"] --> B["🏛️ Domain & State Modeling"]
B --> C["⚡ Feature Engineering & Retrieval Pipeline"]
C --> D["🤖 Agent Orchestration / Decision Model"]
D --> E["🛡️ Guardrails, Red-Teaming & Evaluation"]
E --> F["🚀 Production Serving & Streaming Layer"]
F --> G["📊 Observability, Health & Human Escalation"]
style A fill:#1e2327,stroke:#4c72b0,stroke-width:2px,color:#ffffff
style B fill:#1e2327,stroke:#4c72b0,stroke-width:2px,color:#ffffff
style C fill:#1e2327,stroke:#4c72b0,stroke-width:2px,color:#ffffff
style D fill:#1e2327,stroke:#4c72b0,stroke-width:2px,color:#ffffff
style E fill:#1e2327,stroke:#4c72b0,stroke-width:2px,color:#ffffff
style F fill:#1e2327,stroke:#4c72b0,stroke-width:2px,color:#ffffff
style G fill:#1e2327,stroke:#4c72b0,stroke-width:2px,color:#ffffff
flowchart TD
T["💳 Raw Transactions"] --> SA["🔗 Shared Attribute Extraction"]
SA --> BG["🕸️ Bipartite Graph Construction"]
BG --> GP["📐 Graph Projection"]
GP --> LC["👥 Louvain Community Partitioning"]
LC --> GF["🧬 Graph Topology Features"]
GF --> ML["⚡ LightGBM Decision Engine"]
style T fill:#1e2327,stroke:#4c72b0,stroke-width:2px,color:#ffffff
style SA fill:#1e2327,stroke:#4c72b0,stroke-width:2px,color:#ffffff
style BG fill:#1e2327,stroke:#4c72b0,stroke-width:2px,color:#ffffff
style GP fill:#1e2327,stroke:#4c72b0,stroke-width:2px,color:#ffffff
style LC fill:#1e2327,stroke:#4c72b0,stroke-width:2px,color:#ffffff
style GF fill:#1e2327,stroke:#4c72b0,stroke-width:2px,color:#ffffff
style ML fill:#1e2327,stroke:#4c72b0,stroke-width:2px,color:#ffffff
Areas I work in:
- 🤖 Agentic AI & multi-agent orchestration
- 🔎 Retrieval-augmented generation (RAG)
- 🛡️ AI security & adversarial robustness
- 💳 Fraud detection & credit risk modeling
- 📈 Forecasting & dynamic pricing
- 🎧 Support automation & ticket triage
- 🧾 Contract & document intelligence
- 🕵️ Codebase intelligence & automated review
- ⚙️ AIOps, monitoring & incident response
- 🧬 Feature engineering & ML data infrastructure
🧩 flowmindAGENT ORCHESTRATION Composable agentic workflow engine built on the Actor model — mailboxes, DAG scheduling, and compensation sagas for workflows that need to unwind cleanly when a step fails. |
🗣️ querymindAGENTIC AI Natural-language SQL analytics agent — schema introspection, an AST compiler pipeline, enforced read-only execution, and a self-correcting REPL loop when a query fails. |
🛡️ ragshieldAI SECURITY Agentic RAG security research framework — an intentionally vulnerable pipeline covering attack vectors like corpus poisoning and tool injection, paired with modular defenses for red-teaming and benchmark development. |
REAL-TIME FRAUD DEFENSE Real-time credit card fraud defense — CQRS and event sourcing for auditability, stream-safe rolling state for live scoring, and an interactive TUI for fraud analysts. |
|
CREDIT RISK Bank-grade credit risk scorecard engine — Weight-of-Evidence and Information Value binning, with automated adverse-action reason codes for regulatory-grade explainability. |
⚡ wattcastFORECASTING Production-grade short-term energy load forecasting — 24-horizon direct models, weather coupling, and peak-demand alerting for grid operations. |
|
Autonomous agents that coordinate tasks, execute tools, and recover from failures.
|
High-precision retrieval systems that cite verifiable sources and prevent hallucination.
|
|
Hardening generative AI against adversarial exploits, data leaks, and poisoning. |
Real-time transaction scoring and bank-grade credit risk modeling.
|
|
Predictive optimization for revenue, energy grids, and inventory.
|
Autonomous SRE incident triage, metric anomaly detection, and automated review.
|
|
Production ML pipelines, high-throughput feature stores, and predictive fleet health.
|
|
| Pattern | Value & Focus | Reference Repo |
|---|---|---|
| Actor Model / DAG Scheduling | Resilient async orchestration & saga rollback | flowmind |
| Self-Correction REPL | Natural-language query AST compilation loop | querymind |
| Adversarial Red-Teaming | Poison injection & retrieval defense evaluation | ragshield |
| CQRS & Event Sourcing | Audit-grade stream state & zero data loss | fraudshield |
| Weight-of-Evidence Binning | Regulatory-compliant credit risk scoring | riskscore |
| Direct Multi-Horizon Forecast | Peak load modeling with weather coupling | wattcast |
| Hexagonal Architecture | Port/adapter isolation for enterprise RAG | knowbase |
| Corrective Retrieval | Fallback web augmentation on low confidence | corrector |
| Differential Privacy | Strict mathematical privacy guarantees | privgen |
| Finite State Machine | Deterministic incident diagnosis & actions | alertpulse |
| Point-in-Time Correctness | Zero-leakage distributed feature serving | featforge |
|
Anyone can build a demo that succeeds under happy-path conditions. Real engineering is building software that continues to operate reliably when inputs get noisy, models hallucinate, dependencies fail, or adversaries probe boundaries. |
|
🛡️ Graceful Failure Automated fallback, escalation & saga compensation |
🔍 Full Auditability Immutable event logs, CQRS & decision provenance |
⚖️ Model Calibration Probabilities you can actually bet capital on |
👤 Human Oversight Confidence gating with smooth operator override |
Agentic systems and decision models that hold up under load.
© Nathaniel Gordon · AI / ML Engineering





