Madhavrao

Agentic AI · architecture · data · operations

I help teams move agentic AI from demo to production.

The hard part is rarely the model. It is the architecture around it: memory, state, tools, data, evaluation, recovery, and ownership. I design those seams so the system can be trusted when the happy path ends.

See how I can help See the reference builds → Explore the garden →
Portrait of Madhavrao

Current focus
Production-grade agentic AI reference systems, memory, evaluation, and governed change.

11 reference builds 20+ years across the stack 3 ways to engage 1 production question at a time

What I help teams do

Make the seams explicit.

01

Make the production decision

Reconstruct what an agent actually does, test consequential trajectories, and identify what must change before the next release.

Architecture review → 02

Make execution recoverable

Design checkpoints, cancellation, idempotent effects, replay, and operational evidence for work that outlives a process.

See durable execution → 03

Give agents something true to work with

Build typed memory and data foundations with provenance, scope, versioning, retention, and useful retrieval semantics.

See memory over a lakehouse →

More proof

Built around production seams.

Browse the work →
Agent sandbox boundary diagramPlatform engineeringAgent sandbox sized to blast radiusIsolation, capability-granted mounts, egress controls, and reconstructable audit trails. Agent evaluation and observability architectureAgentic AIEvaluation and observability harnessA typed run ledger for traces, replay fixtures, eval gates, drift signals, and cost attribution. Sutra Feed Guard data governance interfaceData platformSutra Feed GuardDeterministic drift classification and governed change before production data is touched.

Current portfolio

A growing set of production seams.

Eleven reference builds currently cover the production seams around agentic AI: memory, evaluation, privacy, routing, execution, data foundations, exception handling, sandboxing, model control, and governance. New use cases can be added as the work evolves.

Review all 11 builds →

Selected garden notes

Four arguments from the production edge.

Browse all notes →
Agent permissions should expire like leasesAuthority → scope → delegation → expiry Telemetry should preserve decisionsCandidates → commitments → replay → attribution Data residency before cheap inferencePlacement → state → provenance → policy Webhooks are distributed transactionsOutbox → retries → idempotency → compensation

The garden

Start with a question, not an archive.

Browse all notes →
Moving from demo to productionArchitecture → observability → reliability → governance Designing agent memorySchema → provenance → retrieval → forgetting Understanding agent behaviorRuns → trajectories → replay → evidence

Have a production question?

Tell me what you are building, where it is stuck, and what “done” means.

Start the conversation