Manish Pande Menu
Book a consultation
© 2026 Manish Pande Mumbai, India Set in Space Grotesk · Source Serif 4 · JetBrains Mono

AI platform practice

Production agentic-AI platforms, from first principles.

Architecture as a continuous, evolving discipline — from design through implementation. I scale enterprise agentic-AI platforms and the teams behind them from 0 to N, so autonomous agents run reliably, safely, and economically — infrastructure, governance, and scale.

At Jio Platforms I architected and led the 0→production build of the company's enterprise agentic-AI platform and directed a 54-person AI/ML organization; five teams shipped 7 production agents on shared, governed infrastructure, including Jio's first production MCP-based agent. Now Fractional CTO and architect for health-tech and fintech.

Book a consultation

A short call on your agentic-AI platform.

View the system design

8 parts · 30 chapters.

Field guide · 8 parts · 30 chapters

The Agentic Platform

A first-principles system design for running autonomous AI agents reliably, safely, and economically at scale.

Reference architecture View the system design →

Field manual · the 12-stage agentic SDLC

The AIDLC Field Manual

Running the AI Development Lifecycle end to end — spec to self-healing — with harness and loop engineering as the enablers and gates that double as SOC 2 evidence.

Field manual Read the field manual →

The practice

Four kinds of engagement, each grounded in work I have actually shipped or authored — not a service menu.

02 Agents in production Health-tech · fintech
03 The agentic lifecycle Spec → self-healing
01 Agentic-AI platform
LLM serving Data & memory Orchestration LLMOps App hosting
04 Security · Governance · Compliance · Audit Designed in, not bolted on
One platform: governance underneath, agents and the lifecycle on top — four engagements, one system.
  1. 01 · Agentic-AI platform architecture Layered architecture stack, defined end to end for the enterprise

    The five-layer stack named in The Agentic Platform: LLM serving → data & memory → orchestration → LLMOps / observability → app hosting — with security, guardrails, and compliance embedded at the platform layer, not bolted on. The architecture I built 0→production at Jio.

    Architecture · stack selection
  2. 02 · Agents into production Standing up the infrastructure, then shipping the agents on it

    Health-tech: an ontology-driven advanced-RAG agent for multi-hop, evidence-based clinical Q&A. Fintech: a generative-UI agentic app (CopilotKit) for merchant and customer onboarding, loyalty, and analytics.

    Build · embedded or async
  3. 03 · The AI development lifecycle, end to end Standing up the agentic SDLC — from spec to self-healing

    A twelve-stage workflow where humans steer with specs and gates, fleets of agents implement in isolated sandboxes, and an agentic-SRE loop feeds fixes back into implementation — with harness and loop engineering as the enablers, and the review and security gates doubling as SOC 2 evidence. The full method is set out in the AIDLC Field Manual.

    Workflow · harness & loop engineering
  4. 04 · Privacy-aware AI memory & compliance Compliance designed into the agent, not audited after it

    DPDP and GDPR Article 17 right-to-be-forgotten across knowledge graphs, embeddings, and inference logs; regulator-verifiable signed deletion receipts; and multi-hop audit trails.

    Governance · DPDP / GDPR

Writing

I write to think. Technical work on agentic-AI systems, platform architecture, and the governance decisions that shape how teams ship agents in production.

More writing →

About

I architect enterprise agentic-AI systems — the infrastructure, governance, and scaling patterns that take autonomous agents to production. Thirteen-plus years building the systems other teams build on, from India's largest data lake and machine learning at national scale to Jio's enterprise agentic-AI platform.

Read the full background →

Contact

Selective — taking a small number of agentic-AI platform engagements. Email is the fastest path.

Book a consultation

Reply within a couple of days.