AI system architecture
Service boundaries, model integration, APIs, and operating models that connect a business problem to a system a team can run.
Principal AI/ML Architect · Bengaluru, India
Designing AI systems
that survive
production.
I work at the intersection of AI architecture and hands-on engineering—designing agents, retrieval, evaluation, and enterprise systems that operate reliably in production.
From strategic architecture and system design to implementation, security, governance, and long-term adoption.
- 01Foundation modelsCapability
- 02Agents & orchestrationCoordination
- 03Retrieval · tools · dataContext
- 04Evaluation & observabilityEvidence
- 05Security & governanceTrust
Production systems
Designed together. Operated as a whole.
STRATEGY ↔ ARCHITECTURE ↔ ENGINEERING
01 / AREAS OF PRACTICE
Across the system.
Agents & orchestration
Tool use, state, identity, authorization, and recovery paths for workflows with model-driven decisions.
Retrieval & knowledge
Ingestion, indexing, hybrid retrieval, ranking, and freshness—built around the information a workflow actually needs.
Evaluation & reliability
Repeatable datasets, regression checks, human review, and production signals that make quality visible.
Enterprise AI platforms
Shared foundations for identity, data governance, integrations, and clear organizational ownership.
Production engineering
Latency budgets, observability, deployment, cost control, and maintainable infrastructure beyond the prototype.
02 / SELECTED WORK
Architecture, made concrete.
Public engineering projects. The problems, boundaries, and decisions behind the code.
01Public open-source project
agent-guard
Making every tool call an authorization decision.
A policy boundary between what an agent proposes and what it is permitted to execute.
Agent security · Policy · Auditability02Public open-source project
promptsheon
Treating agent configuration as a release artifact.
Versioned configuration, evaluation, and approval belong in the same path to release.
Evaluation · Release engineering · Agent workflows03Public research implementation
delta-search
Separating a search strategy from the graph problem.
A reusable reward–penalty framework for heuristic subgraph extraction in Python.
Graph algorithms · Incremental computation · Research engineering03 / HOW I THINK
Architecture is more
than choosing models.
I approach AI as a systems problem. Model capability matters, but production outcomes depend just as much on data, evaluation, security, observability, cost, and operational ownership.
The boundaries between those concerns are where much of the engineering happens.
Read more about my approach04 / PUBLIC ENGINEERING
Open-source engineering.
TypeScript
agent-passport
Identity, delegation, and trust for autonomous AI agents.
Go
alfred
Local-first research agent runtime: HTTP/SSE turn loop, SQLite+JSONL persistence, 25+ MCP workers (search, plotting, evidence DAG, image generation).
TypeScript
revex
Hybrid retrieval (vector + BM25 + graph) for TypeScript.
05 / WRITING
Thinking through the details.
Designing reliable AI agents starts at the tool boundary
Identity, authorization, durable state, and recovery determine what an agent can safely do after it chooses an action.
Evaluation is infrastructure, not a release-day checklist
A useful evaluation system connects versioned evidence, explicit failure categories, and release decisions across the life of an AI application.
A GOOD CONVERSATION STARTS SOMEWHERE
Let’s connect.
Working on a difficult AI or engineering problem? Hiring for a senior architecture role? Or simply interested in exchanging ideas?
I’d like to hear from youBengaluru, India