Sagar Trivedi · Applied AI Architect

Summary

I have spent more than fourteen years across software, infrastructure and platform engineering. For the last several I have been the technical counterpart stakeholders work with when an AI idea has to become something a business can actually run: discovery and use-case selection, architecture across models, data, integration, security and governance, then the evaluation and telemetry that show whether it works. I lead that work now at Honest, and I did the client-facing version of it at Thoughtworks, embedded with engineering leaders. I also build tooling that puts AI capability directly in the hands of people who do not write code.

Experience

Honest

Current

Senior DevOps Engineer, AI & Platform

Mar 2024 – Present · Bangkok, Thailand

I lead three concurrent initiatives: AI enablement and implementation, high-availability engineering, and PCI DSS certification.

AI delivery and adoption

Lead the technical side of an AI initiative delivering a customer-care copilot, partnering across product, operations, engineering, risk, compliance and quality control to move it from prototype into daily use.
Designed the phased path from human evaluation to a human-in-the-loop production workflow, spanning streaming ingestion, warehousing, retrieval over customer and knowledge context, model serving and monitoring.
Built and operate the evaluation stack: an LLM-as-judge harness scoring response grounding, knowledge-base synchronisation, model-comparison runs, dashboards and recurring audits.
Calibrated the judge against human review, reaching 94–97% precision on flagged cases across two audits.
Ran a paired golden-set evaluation with both arms regenerated and McNemar significance testing, which supported a model change that reduced hallucination rate by roughly 28% relative and cut response verbosity by 27%.
Instrumented quality, latency, tokens, retries and cost per request through Prometheus and Grafana, with model-aware attribution.
Designed a prompt and context caching strategy that serves 98.5% of input tokens from cache, bringing cost per request down to a fraction of a cent, roughly a 7× reduction against the uncached baseline.
Introduced cost per prevented error as the decision metric for model selection, replacing cost per token.

Platform and reliability

Run Kubernetes and GitOps delivery across AWS and GCP with Terraform, Helm and Argo CD.
Held 99.965% uptime over a 90-day window, with buffered ingestion and reliable recovery after restarts.
Own FinOps across rightsizing, commitment planning and cost attribution, cutting annual cloud spend substantially across both clouds.
Lead high-availability engineering and the PCI DSS certification workstream in a regulated environment.

Thoughtworks

Dec 2021 – Feb 2024 · India

Lead Infrastructure Consultant Aug 2023 – Feb 2024
Senior Infrastructure Consultant Dec 2021 – Aug 2023

Embedded with client engineering teams as their primary technical counterpart, translating ambiguous business priorities into cloud, network and platform architectures that fit legacy, regulatory and organisational constraints.
Headed a platform team, set the roadmap and architecture standards, and mentored architects and senior engineers while staying hands-on in delivery.
Partnered with a client Head of Engineering on a transformation strategy covering technical debt, security and reliability, and introduced FinOps practices that reduced their cloud spend.
Wrote cost-optimisation positions for executive audiences and ran workshops that turned technical guidance into repeatable capability the client could carry on themselves.
As Senior Infrastructure Consultant, established the cloud and infrastructure practice, standardised CI/CD patterns, and built long-term platform partnerships with senior client stakeholders.

Cactus Communications

May 2014 – Dec 2021 · 7 years · Mumbai, India

DevOps Architect Jul 2020 – Dec 2021
DevOps Engineer Aug 2019 – Jul 2020
Software Architect Jun 2016 – Aug 2019
Software Developer May 2014 – May 2016

Owned DevOps and security operations for multiple business units, translated product requirements into platform strategy, and led an AWS monolith-to-microservices transition using Terraform and Jenkins.
Built automated delivery platforms and operated secure, high-availability microservices across AWS networking, compute, containers, messaging, APIs, databases, storage, caching and analytics.
Owned application architecture and migration strategy earlier in the role, partnering with product teams on roadmap decisions and mentoring the engineering team.
Delivered revenue-critical enterprise web applications and integrations, owning maintainability, reliability, security, availability and performance.

Tech5 Solutions

Chief Technical Analyst

Apr 2011 – Apr 2014 · Mumbai, India

Translated client business needs into functional designs and worked with development teams to deliver systems on time and within budget.

Open Source

Projects I build and maintain.

SprintFoundry TypeScript · multi-agent delivery

An orchestrator that turns GitHub, Linear and Jira tickets into tested, reviewed pull requests through specialised product, architecture, development, QA, security and DevOps agents. Runs on Claude Agent SDK and OpenAI Codex runtimes with sandboxed execution, explicit planning, deterministic validation, human review gates, Kubernetes deployment, live monitoring, and cost and runtime visibility through OpenTelemetry.

TriSeek Rust · code search MCP server

A local-first, read-only MCP server with a trigram-indexed search engine measured up to 16.9× faster than ripgrep on large repositories. The tool surface is deliberately bounded to search and context, with no file editing and no shell execution. A session memo layer, validated across replayed sessions with no false negatives, cuts redundant context reads across coding-agent workflows, and context is portable across Claude Code, Codex, OpenCode and Pi.

Skill Miner AI enablement

Mines past agent sessions to generate reusable, shareable team skills. The resulting skills are in use by non-engineering teams.

OptiPod Kubernetes operator

GitOps-safe Kubernetes resource rightsizing operator. Resolves the conflict where vertical autoscaling and Argo CD fight over resource definitions.

claude-k8s-plugin Developer tooling

Claude Code plugin for scaffolding Kubernetes operators and admission webhooks.

Skills

AI systems: LLM evaluation and judging, human-calibrated review, agent orchestration, MCP servers, prompt and context design, caching and cost optimisation, retrieval design, guardrails
Architecture: APIs, distributed systems, data integration, event streaming, identity, security, privacy
Platform & cloud: Kubernetes, Docker, Argo CD, Terraform, Helm, GitOps, Jenkins, AWS, GCP, Vertex AI
Observability & reliability: Prometheus, Grafana, OpenTelemetry, SLOs, incident response, high availability
Languages: Python, Go, Rust, TypeScript, Bash
Working with customers: Executive and engineering discovery, use-case prioritisation, architecture roadmaps, prototypes and proofs of value, technical workshops, enablement and mentoring, FinOps, regulated-environment compliance

Writing

I write about the parts of AI systems that decide whether they survive production: evaluation, failure modes, cost, and the protocols and tooling underneath.

Agents, evaluation and retrieval

Model Context Protocol

Inside Codex CLI, an eight-part series

Agent tooling and team practice

All 25 articles

Certification & Education

Certification: AWS Certified Solutions Architect – Professional
Education: B.E., Information Technology, Mumbai University, 2005–2008
Diploma, Computer Technology, 2002–2005
Languages: English, Hindi, Marathi