Building the only stack where a domain-specialist AI model can make decisions a regulator can audit.

I work on the seam between what AI can do and what institutions can trust. The work is the integration of three layers into one stack:

• Reasoning layer — domain-specialist models I am fine-tuning on Gemma 4. Solen for supply chain, Verac for finance and settlement, Axiom for financial markets. Specialists trained to reason like a 30-year veteran in their field. Generalist models are wide; these are deep, and the gap widens as the stakes go up.
• Substrate layer — the open infrastructure primitives the models run on. Persistent memory the agent owns. Cryptographic identity. Runtime policy enforcement. Temporal reasoning. Intent verification. Everything an AI needs to be remembered, signed, and held accountable.
• Settlement layer — XAP Protocol and the Verity truth engine. When a decision involves value, every settlement is deterministically replayable by a third party who was not in the room.

A model alone is a smart oracle. The substrate alone is interesting open-source primitives. The settlement layer alone is a payment protocol. Together they are the only stack where a fine-tuned domain specialist can make decisions you can prove three years later, in a courtroom, without the original engineer present. That is the work.

Published research: 100+ citations across ethical AI in financial decisioning, credit risk modeling, supply chain finance, and systemic risk.

→ omoshola.me

• 🏗️ I design complex AI systems as composed layers: data contracts, model services, reasoning engines, policy controls, and observable execution paths.
• 🔁 I prioritize architecture-level guarantees: traceability, determinism where required, graceful degradation, and explicit failure boundaries.
• 🧭 I treat governance as a system primitive, not an afterthought: explainability, audit logs, access control, and policy enforcement are built into core workflows.
• ⚙️ I enjoy hard systems problems at scale: multi-agent orchestration, graph-native memory, temporal reasoning, and reliability under real-world constraints.

Two organizations carry the work: Agentra Labs (open agentic infrastructure) and Agentra Commerce (XAP, Verity, ZexRail — production agent settlement).

• 🧬 agentralabs-models — Domain-specific model training pipeline. Fine-tuning Gemma 4 to think like world-class domain experts, not generalists. Reasoning-first training across six categories that teach how experts think, not just what they know. Quality-scored data (4+/5 on reasoning depth, domain accuracy, calibration, practical value) and failure-pattern training on expert mistakes and corrections.
  • Solen — Supply Chain Management (training)
  • Verac — Finance / Settlement (pipeline ready)
  • Axiom — Financial Markets (pipeline ready)

• 🐉 Hydra — The AI that remembers you. Forever. A living digital entity built in Rust. 68 crates. Self-writing genome. Persistent memory. Constitutional governance. Drop a TOML file, Hydra learns. The thesis is that the next AI is not a model you call — it is an entity that grows alongside you, with memory you own and a constitution you can audit.
• 🦀 Nexus — Supply chain intelligence platform in Rust. Treats demand, lead times, and supplier reliability as probability distributions. Monte Carlo simulation, temporal graph kernel, agentic recommendations a procurement team can audit. (Powered by Solen.)

• 🧱 XAP Protocol — eXchange Agent Protocol. The open economic protocol for autonomous agents: identity, negotiation, conditional escrow, execution receipts, and deterministic decision replay.
• 📦 xap-sdk — Python SDK for XAP. Agent discovery, negotiation, settlement, Verity receipts, and 8 MCP tools for Claude and Cursor. pip install xap-sdk.
• ✅ Verity Engine — The open-source truth engine for XAP. Deterministic replay, hash chains, RFC 3161 timestamps, and seven trust properties. (Powered by Verac.)
• 🚆 ZexRail — Production XAP infrastructure. Rust microservices, PostgreSQL, Stripe Connect integration in progress, Verity integration. The backend is a private repo; the live site is zexrail.com. (Powered by Verac.)
• 🔭 Verity Observatory — Live public observatory for the truth engine. Paste any receipt hash and verify it independently. No account required.

The Agentra sisters. Each one is a standalone Rust core + MCP server, named for the cognitive primitive it gives an agent.

• 🧠 AgenticMemory — Persistent cognitive graph memory. Facts, decisions, reasoning chains, corrections. 16 query types, sub-millisecond. Rust core + Python SDK + MCP server.
• 👁️ AgenticVision — Persistent visual memory. Capture screenshots, embed with CLIP ViT-B/32, compare, recall.
• 🧩 AgenticCodebase — Semantic code intelligence. Compile repositories into navigable concept graphs with impact analysis, coupling detection, and prophecy.
• 🪪 AgenticIdentity — Cryptographic agent identity. Ed25519 anchors, signed action receipts, scoped trust delegation. One .aid file.
• ⏱️ AgenticTime — Temporal reasoning. Deadlines, schedules, sequences, duration estimation (PERT), decay models. One .atime file.
• 📜 AgenticContract — Policy engine. Enforceable rules, risk limits, approval gates, obligation tracking, violation detection. One .acon file.
• 📡 AgenticComm — Structured agent-to-agent and agent-to-human communication. Channels, pub/sub, routing, presence. One .acomm file.
• 🎯 AgenticPlanning — Persistent intention infrastructure. Goals, decisions, commitments, strategic reasoning. .aplan file format.
• 🌐 AgenticData — Universal data comprehension. Infer schemas, track lineage, detect anomalies, transform any format. 122 MCP tools, 16 parsers.
• 🔄 AgenticWorkflow — Universal orchestration engine. Workflows, pipelines, state machines, batch processing. 24 inventions, 124 MCP tools, .awf format.
• 🔌 AgenticConnect — Universal external interface engine. 123 MCP tools, 18 protocols, Connection Souls, Intelligent Retry, Encrypted Vault.
• 🧭 AgenticVeritas — Intent compilation and uncertainty detection. Truth verification, ambiguity resolution, causal reasoning.
• 🧬 AgenticCognition — Longitudinal user modeling. Living models of human consciousness for AI agents.
• 🌍 AgenticReality — Existential grounding. Deployment awareness, resource proprioception, reality physics.

Platform layer — agentic-evolve (pattern-library engine for instant rebuilds), agentic-aegis (streaming validation + shadow execution), agentic-forge (project-blueprint engine), agentic-sdk (shared traits and contracts across all sisters), and agentralabs-tech (workspace orchestrator).

• 🏦 Explainable credit risk intelligence — Architected an explainable ML credit risk pipeline for 200,000+ applications using ensemble modeling + real-time macroeconomic feature integration, with transparent reasoning outputs for underwriting; reduced default rates by 15% while expanding access to underserved populations.
• 🔐 Privacy-preserving synthetic data — Built a synthetic data generation stack combining Gaussian Copula and GAN-based synthesis for secure AI model development, achieving full anonymization with regulatory alignment while preserving statistical utility for downstream risk prediction.
• 📦 Supply chain resilience modeling — Developed forecasting and risk pipelines using ARIMA, neural time-series models, and ML supplier-risk scoring to detect disruptions early; reduced stockouts by 22% and drove $2M in cost savings via model-informed operations.

• 🔬 Explainable Credit Intelligence — SHAP-based interpretable risk scoring (CrossSHAP methodology)
• 🌐 Network Analysis — Supply Chain — Systemic risk assessment in supply chains
• 🏦 Credit Risk Creditum — Credit risk modeling and analysis
• 📊 RiskX — Risk analytics engine
• 📄 AgenticMemory Paper I — Binary format and memory substrate design
• 📄 AgenticMemory Paper II — Query expansion and retrieval strategy
• 📄 AgenticMemory Paper III — MCP server architecture and tooling
• 📄 AgenticVision Paper I — Visual/web graph representation and reasoning
• 📄 AgenticVision Paper II — Vision MCP server design
• 📄 AgenticCodebase Paper — Semantic compiler and code intelligence architecture
• 📄 AgenticIdentity Paper — Trust anchors, identity proofs, and receipt model
• 📄 AgenticTime Paper — Temporal format and reasoning model
• 📄 AgenticContract Paper — Policy engine and governance constraints
• 📄 AgenticComm Paper — Agent communication runtime and protocol design
• 📄 IEEE Academic Reviewer — Peer reviewer for ICAD 2025
• 📐 IEEE Standards Development — AI ethics, cybersecurity, financial LLM requirements, supply chain security

• 🏅 IEEE Senior Member, 2025 — Elevated in recognition of significant contributions to the profession; eligible for executive volunteer positions and review panel service. Active memberships: IEEE Computational Intelligence Society, IEEE Consumer Technology Society, IEEE Technology and Engineering Management Society, IEEE Young Professionals.
• 🎯 Ethics and Conference Reviewing — NeurIPS 2025 (Datasets & Benchmarks), DeepLearningIndaba 2025, IEEE ICMI 2026 (King Faisal University), IEOM 2025 World Congress (University of Windsor), IEEE IATMSI, and 2025 International Conference on Signal Processing, Computation, Electronics, Power and Telecommunication.
• ⚖️ Judging and Evaluation — TrackShift Innovation Challenge 2025 (Mphasis F1 Foundation x MoneyGram Haas F1 Team), HackNC 2025 (UNC Chapel Hill), ASA Statistics Project Competition (Grades 7-12; 30+ projects), and ASA USCLAP.
• 🧪 Journal Peer Review — Journal of Data Analysis and Information Processing (JDAIP), including LLM-powered enterprise intelligence, healthcare big data, and cloud optimization work.
• 🤝 Mentorship — SciPy Conference 2025 Mentorship Program and Nova Talent Elite Mentorship Program, supporting emerging and senior professionals in ethical AI, financial AI, and supply chain analytics.

• 🏛️ OSTP Federal AI Policy Contributor (2025) — Submitted technical recommendations on AI regulatory reform to address SR 11-7/SR 23-4 barriers to explainable financial AI, including SHAP/LIME recognition, federal data API clarification, and interagency coordination; quantified a 10-15% potential credit-loss reduction impact across the $18.04T household debt market. (Notice 90 FR 46422 • Docket OSTP-TECH-2025-0067 • Submission)
• 🌐 U.S. Commerce Federal AI Export Strategy Contributor (2025) — Submitted strategy guidance for the American AI Exports Program focused on compliance-native, explainable, and secure AI exports (Basel III/IV, GDPR alignment, SHAP/LIME, IEEE harmonization), with emphasis on financial resilience and allied infrastructure development. (Notice 90 FR 48726 • Docket ITA-2025-0070 • Submission)

• 🧬 Specialist model family (Solen / Verac / Axiom) — Fine-tuning Gemma 4 into reasoning-first domain experts for supply chain, settlement, and financial markets. The thesis: in regulated industries, depth beats breadth, and a model trained to reason like a specialist outperforms a generalist on the work that actually matters.
• 🦀 Nexus — Shipping a supply chain OS in Rust with probabilistic planning, simulation-first decision support, and auditable agent recommendations.
• 💳 FyxCred — Building cashflow-native credit intelligence for credit-invisible populations with explainable scoring, policy-aware decisions, and governance-ready outputs.
• 🧠 Agentra Sisters — Advancing MCP-native, artifact-portable infrastructure across graph memory, multimodal vision, semantic code intelligence, identity, time, policy, and communication.
• 🧪 Applied AI R&D — Developing explainability, uncertainty-aware modeling, and privacy-preserving data methods for regulated production systems.
• 📝 Research Service — Reviewing AI and cybersecurity research through IEEE and related scholarly programs.

• How to engineer trustworthy agent systems over time, not just at launch.
• How to operationalize SHAP/LIME-style explainability so adverse-action reasons satisfy ECOA/FCRA requirements in real workflows.
• How to build credit models that expand access for people historically excluded by formal scoring systems.
• How to combine uncertainty quantification, calibration, and drift monitoring into continuous model governance.
• What memory, identity, policy, and reasoning primitives are required before agents can safely operate in regulated domains.
• How to use privacy-preserving synthetic data to balance data utility, security, and compliance.
• How African knowledge systems can inform modern computation and AI system design.

• Modeling: explainable ML, ensemble risk models, time-series forecasting (ARIMA + neural), supplier-risk scoring.
• AI Safety & Governance: model transparency, adverse-action traceability, policy-aware decision systems, regulatory-compliant AI deployment.
• Data & Privacy: synthetic data generation (Gaussian Copula + GAN), anonymization, utility-preserving data pipelines.
• Agent Infrastructure: graph-native memory, multimodal retrieval, identity and trust primitives, temporal reasoning, contract-constrained actions.