Founder of Parmana Systems. I build controls that let AI agents act on real systems: every action is checked against signed policy, and every record can be verified offline by anyone.
I am a technology entrepreneur, AI systems builder and independent researcher. My work covers decision infrastructure, autonomous systems and verifiable computing, with a focus on how organisations keep control over actions taken by AI, software and people.
At Parmana Systems I am developing deterministic authorisation, execution integrity and cryptographically verifiable evidence. The aim is to turn business intent into enforceable rules, so that authorisation is explicit, execution is predictable and outcomes can be verified independently.
What can an AI agent do, and how do we prove it?
A prompt instruction is not a control. What holds up is a check before each action that a person has approved, and a record a third party can verify without trusting the system that produced it.
The core (parmana-core) and the verifier are open source under the Apache 2.0 licence.
Reported two advisories in AiSOC, an open-source AI security operations platform: GHSA-ghpm-vc6j-hp5r and GHSA-hrc5-w9r3-2g8r, both fixed in PR #1750. Listed in AiSOC's credits.
- Why We Are Testing Whether Our Tests Can Be Trusted (engineering essay, November 2026). It argues that if a system claims unauthorized actions cannot execute, we must test whether our tests would detect the mechanisms that make that claim false. It draws on mutation testing of Parmana's cryptographic core.
- The Infrastructure Dependency Gap: Accountable Execution in Autonomous AI Systems Under Infrastructure Failure (position and framework paper, November 2026). It argues that autonomous AI systems must keep business authority and execution accountability when infrastructure fails, and proposes a deterministic architecture for doing so. Experimental validation is pending.
- The Authority Gap: Why Autonomous AI Needs Deterministic Execution Boundaries (discussion draft, November 2026). It argues that an institution's authority over an AI agent must be declared independently, decided deterministically and enforced at the point of execution, with the evidence kept verifiable. The paper is available as a PDF in the
research/folder of the parmana-core repository. - Beyond Trust But Verify: Why Autonomous AI Needs Verifiable Execution (research paper, pull request, November 2026). It argues that authority over autonomous AI must be explicit, independently evaluated and verifiable, and cites the Carnegie report and California's frontier AI laws.
- From Machine-Readable Policy to Enforced Authority: Why Autonomous Systems Need Deterministic Policy Enforcement (research paper, November 2026). It proposes a reference architecture that separates AI-generated intent from institutional authority, evaluates authorization independently of the agent, enforces it at the execution gateway and records verifiable evidence.
Published work on Parmana
- Parmana research paper (PDF): the design of Parmana and its approach to verifiable agent authorisation.
- Regulation mapping: how Parmana supports human oversight and record keeping under the EU AI Act, NIST AI RMF, ISO/IEC 42001 and RBI guidance.
- OWASP mapping: how Parmana addresses the OWASP Top 10 for LLM and Agentic Applications.
- How Parmana compares: Parmana set against IAM, API gateways and LLM guardrails, including where it is weaker.
- Mutation testing results: how well the tests in parmana-core's security-critical packages catch deliberate changes to the code, with the scores and the surviving changes reviewed.
- Advisory: product strategy, roadmap and go-to-market for AI platforms and regulated products.
- Architecture and design reviews: implementation support for teams building AI systems that act on money or records.
- Research collaboration: verifiable AI, agent authorisation, and governance and audit evidence.
- Speaking: conferences, panels, webinars and internal workshops on AI governance and trust and safety, for business leaders, boards, regulators and students.
- Related work: trust and safety and fraud-prevention reviews, and 0-to-1 product building.
Typical audiences:
- Business leaders deciding what AI agents may do with money, customers and systems of record.
- Security teams that authorise, monitor and audit agent actions.
- Governance and compliance teams that must show evidence to regulators.
- Educational institutions shaping AI policy and responsible-use guidance.
Book a call to talk through one real problem.
- Founder and Product Lead, Parmana Systems (August 2023 to present): built the authorisation and enforcement layer for AI agents, from 0 to 1 as sole builder, covering problem definition, architecture and implementation.
- Product Manager, Bada Business (February to July 2023): launched 0-to-1 products for digital learning and business education.
- Associate Product Manager, Shaadi.com (October 2021 to January 2023): onboarding, activation and trust and safety for a platform with more than 2 million monthly active users.
- Associate Product Manager, MakeMyTrip (April 2018 to March 2020): marketplace and supply-side products across hotels.
- Account Manager, MakeMyTrip (April 2016 to March 2018): strategic partnerships for a hospitality portfolio above ₹1,500 crore a year.
- Founder, 132Connect (April 2015 to March 2016): a platform connecting engineering students with jobs.
- Software Engineer, Tata Consultancy Services (March 2010 to April 2013).
- MBA in Finance, IMT Ghaziabad
- B.Tech, NIT Srinagar
- Mohinder Singh Charak, Co-Founder and Director. Indian Army veteran with 30 years of service.
- K. Venkata Abhigna, Partnership Manager. Works on AI automation for high-precision industries, building zero-error workflows for manufacturing, aerospace, defence and medical devices.
- Consulting, speaking and partnerships: [email protected]
- LinkedIn: linkedin.com/in/pavancharak
- Website: parmanasystems.com
- Calendar: cal.com/pavan-charak