Dr. Ravi Teja Thiriveedhi · clinician turned AI systems architect · I design, ship and govern production multi-agent systems, with 20 years of clinical practice and hospital operations behind every decision. Hyderabad, India · open to remote roles (IST, UTC+5:30).

Start here: WhatsApp agent case study (a live client system) · Agent governance kit · LLM token economics · All showcases

Try the live demos (invented data, run in your browser): Agent governance gate · Owner approval desk · Heartbeat token calculator · Control plane playground

Contact: dr.raviteja1981.new@gmail.com · LinkedIn · Website · private walkthroughs of my own production systems on request, under NDA (client systems only with the client's permission)

For twenty years I ran the operating side of healthcare: I treated patients, founded two healthcare facilities and ran a third as senior administrator, and wrote the SOPs, audits and financial controls that kept them safe and solvent. In 2026 I brought that operator discipline to AI. I now design, ship and govern production multi-agent systems — and make them pay for themselves.

WhatsApp agent case study · live demo	Agent governance kit · live demo
LLM token economics · live demo	Empire OS reference architecture · live demo

flowchart LR
    subgraph T1["Tier 1 · Seat (MacBook M3)"]
        P["Planner<br/>Claude"] --> B["Builders<br/>Claude Code · Codex"]
    end
    subgraph T2["Tier 2 · Control plane (2012 Mac mini, 8 GB)"]
        G["OpenClaw gateway<br/>9 governed agents"] --- GOV["Governor · kill switch<br/>approvals · audit log"]
        G --- OBS["Telemetry · alerts<br/>backups · DR drills"]
    end
    subgraph T4["Tier 4 · Client edge nodes"]
        C["Client platform<br/>15 agents · WhatsApp · Telegram"]
    end
    subgraph T5["Tier 5 · Free cloud"]
        GH["GitHub · encrypted Drive<br/>Tailscale · Telegram"]
    end
    B -- "test-gated releases" --> G
    G -- "one-way observation" --> C
    G --> GH

I practise disciplined vibe coding: I write the spec, the acceptance criteria and the guard rules; AI coding agents write the code; I own architecture, review, testing, release and operations.

flowchart LR
    S["Spec + task card<br/>acceptance criteria"] --> A["AI builder<br/>under guard rules"]
    A --> R["Independent review<br/>agent or adversarial pass"]
    R --> T{"Release gate<br/>all suites green?"}
    T -- yes --> SHIP["Tagged release<br/>+ evidence file"]
    T -- no --> S

Public showcases

• 🧪 sterilefield/empire-os-reference-architecture — Reference architecture: 9 governed AI agents on an 8 GB Mac mini, 99/99-suite release gate, 1,026 tests. Sanitised showcase of Empire OS.
• 🧪 sterilefield/llm-token-economics — Case study + playbook: an AI agent fleet cut heartbeat tokens 12.0M to 1.0M a day (-91%). 18 methods across 5 groups, with measured before/after.
• 🧪 sterilefield/whatsapp-agent-case-study — Anonymised case study: a 15-agent WhatsApp and Telegram operations platform for a real-estate brokerage. Single-writer data spine, human-in-charge policy, −91% heartbeat tokens.
• 🧪 sterilefield/agent-governance-kit — Autonomy levels, risk classes, approval tiers, kill switches and audit for AI agents, mapped to OWASP Agentic Top 10, NIST AI RMF and ISO/IEC 42001, with a healthcare overlay for India.

Latest releases

• 🚀 llm-token-economics v0.2.0 · 2026-10-21
• 🚀 agent-governance-kit v0.2.0 · 2026-10-21
• 🚀 whatsapp-agent-case-study v0.2.0 · 2026-10-21
• 🚀 empire-os-reference-architecture v0.2.0 · 2026-10-21
• 🚀 agent-governance-kit v0.1.0 · 2026-10-14

_{Updated automatically by GitHub Actions.}

LinkedIn · Email · Website · Hyderabad, India · open to remote roles (IST, UTC+5:30)

_{Engineering numbers come from the linked repositories, release notes and case studies. Operational figures are from my own records and are not independently audited. Graphics are generated by this repository's own workflows.}