Enterprise AI & Platform Architecture Leader for regulated financial services. Building production AI systems, governance frameworks, and trust infrastructure.
Published research · book · IETF draft · open-source protocols and tooling
My research focuses on an open problem: multi-agent AI systems have no standard for identity, delegation, or provenance. The work spans four layers:
Research Specifications Implementations Guides
──────── ────────────── ─────────────── ──────
┌─────────────┐ ┌──────────────┐ ┌──────────────────┐ ┌─────────────┐
│ Agent │ │ AIP Spec │ │ aip (Rust+Python)│ │ Agentic AI │
│ Identity │──▶│ IETF Draft │──▶│ PyPI packages │──▶│ for Serious │
│ (arXiv) │ │ │ │ Framework addons │ │ Engineers │
└─────────────┘ └──────────────┘ └──────────────────┘ └─────────────┘
┌─────────────┐ ┌──────────────────┐
│ Provenance │ │ ldp-protocol │
│ Paradox │─────────────────────▶│ (Rust) │
│ + LDP (arXiv)│ │ │
└─────────────┘ └──────────────────┘
┌─────────────┐
│ DCI: │
│ Collective │ Reasoning layer for multi-agent deliberation
│ Reasoning │
└─────────────┘
Each layer solves a different problem: AIP handles who is this agent and what can it do, LDP handles how do agents route and attest provenance, DCI handles how do agents reason together. These ideas flow from protocol research into practical runtimes and developer tooling. The book covers how to build, evaluate, and govern all of it in production.
| Layer | Paper | Link |
|---|---|---|
| Identity | AIP: Verifiable Delegation Across MCP and A2A | arXiv:2603.24775 |
| Provenance | The Provenance Paradox in Multi-Agent LLM Routing | arXiv:2603.18043 |
| Protocol | LDP: Identity-Aware Protocol for Multi-Agent Systems | arXiv:2603.08852 |
| Reasoning | DCI: Structured Collective Reasoning with Typed Acts | arXiv:2603.11781 |
IETF Internet-Draft: draft-prakash-aip-00
Agentic AI for Serious Engineers — When to use agents and when not to. Evaluation, hardening, governance, and the production reality most teams discover too late. Amazon (paperback & Kindle) | Code companion
| Repository | What it solves |
|---|---|
aip |
Verifiable identity and scoped delegation for AI agents across MCP and A2A (Rust + Python, PyPI) |
ldp-protocol |
Identity-aware routing and provenance for multi-agent systems (Rust, crates.io) |
ai-governance-framework |
50+ governance documents across 8 domains and 8 jurisdictions for regulated AI |
enterprise-rag-bench |
RAG patterns benchmarked for enterprise: chunking, retrieval, eval harness, guardrails |
enterprise-genai-platform |
Reference architecture for LLM applications in banking |
19 years in enterprise technology. VP at a global bank leading cross-location AI and platform architecture. Former Chief Scientist at an AI startup. Executive MBA (ISB), M.Tech in Enterprise Analytics (NUS).
sunilprakash.com | Google Scholar | LinkedIn