Building
Autonomous
AI Agents
Agentic AI Engineer with 18+ years engineering production systems at Amazon, PayPal, GitLab, Freshworks, and Circle — now building autonomous LLM systems with the Claude Agent SDK, LangChain/LangGraph, and RAG pipelines. I combine deep systems and DevOps expertise with hands-on production experience in multi-agent orchestration, tool design, and LLM evaluation.
18+ Years
Systems Engineering
About
Hi, I'm Ramya 👋
I'm Ramya Authappan — an Agentic AI Engineer with 18+ years of experience shipping production systems at FAANG, FinTech, and high-growth SaaS companies, including Amazon, PayPal, GitLab, Freshworks, and Circle. Over the last two years I've been deliberately re-specializing: moving from leading quality engineering organizations to hands-on building of agentic AI systems using the Claude Agent SDK, Claude Code MCP, LangChain/LangGraph, and RAG pipelines.
After leading engineering teams for several years and exploring the engineer–manager pendulum concept by Charity Majors, I deliberately returned to hands-on engineering — and doubled down on where I believe the most consequential work is happening today: agentic AI. At Circle Inc.'s engineering org, I built and shipped a Claude Agent SDK–powered Selective Test Execution system that reads PR diffs, reasons about blast radius, and cuts CI cost by 50% — production proof that LLM orchestration, not just prompting, is my core skill.
I am furthering this specialization through a postgraduate program in Forward Deployed Engineering & Applied AI Solutions at FDE Academy (Futurense), expected 2026. The program has translated directly into shipped work: a driver-facing agentic dispatch console built on Amazon Bedrock AgentCore, a LangChain agent that classifies meeting transcripts and files the resulting action items as GitHub issues, and a client discovery case study modeled on real Forward Deployed Engineering engagements.
My 18 years in quality engineering — architecting CI/CD infrastructure handling 100K+ daily test executions at GitLab, cutting Kindle regression time by 95% at Amazon, and rolling out Consumer-Driven Contract testing across 20+ microservices at Freshworks — gave me a verification mindset that's rare in AI engineering: I don't just build agents, I know how to prove they work.
Beyond my technical work, I'm passionate about building inclusive tech communities. As Director of Women Who Code Chennai (2018–2024), I mentored speakers, organized educational meetups, and created pathways for women to advance in technology careers.
FDE Academy
Applied AI / Agentic AI Projects
Hands-on projects from my Forward Deployed Engineering & Applied AI Solutions program.
Shipping agentic systems end-to-end — LLM orchestration, tool design, memory, observability, and client discovery — the same skill set FDE and applied-AI roles require.
SetuHaul — Agentic Freight Dock Scheduling Console
Driver-facing dispatch console · Aug 2026
Built a driver-facing dispatch console for freight dock scheduling that replaces a rules engine with an LLM agent for operational reasoning — turning free-text driver reports of delays and dock-availability questions into grounded scheduling answers instead of plausible-sounding guesses. The LLM is constrained to a single tool that calls back into the backend over HTTP to check real, currently-open slots, so it can report availability but never invent it; booking itself is a separate deterministic REST call with no LLM in the loop. Conversation memory persists in a managed Amazon Bedrock AgentCore memory resource so context survives redeploys and horizontal scaling.
Architecture
Three independently deployed services — Next.js 15 + React 19 on Vercel, FastAPI + SQLite on Render, LangChain agent on Bedrock AgentCore — wired with tightly scoped IAM
Grounding
LLM reasons over live availability via one HTTP tool call; booking stays a separate deterministic REST call — reasoning is AI-assisted, state changes are not
Observability
LangSmith traces every prompt, tool call, and reply; CloudWatch traces the deterministic booking path too — debugging relies on real execution traces
Meeting Action-Item → GitHub Issue Agent
LangChain agent · Jul 2026
Built a LangChain agent that reads a meeting transcript, classifies each action item as confirmed, tentative, or a suggestion, and files a GitHub issue only for confirmed commitments — skipping tentative items and mere suggestions. Designed a single atomic tool call (create-issue + database update together) to eliminate a race condition where the model could batch tool calls out of order; the agent loops until resolved, then emails a summary table of created and skipped items via Gmail SMTP.
Classification
Every action item triaged as confirmed / tentative / suggestion — only confirmed commitments become GitHub issues
Concurrency fix
Atomic create-issue + database-update tool call eliminates a race condition from out-of-order batched tool calls
Output
Summary table of created and skipped items emailed via Gmail SMTP; state tracked in SQLite
ShipRight Logistics — Client Discovery & Solution Design Case Study
Jul 2026
Produced stakeholder mapping, problem definition, and RAPID decision documentation for a fictional enterprise logistics client — simulating the discovery-to-recommendation workflow used in real FDE engagements.
Circle · Production
Agentic AI in Production
Autonomous, self-healing test systems running in production — not demos.
Deployed directly into Circle Inc's engineering org, I architected an AI-driven quality platform powered by the Claude Agent SDK, Claude Code MCP, and Buildkite MCP. Each tool below is production-shipped and running at Circle today.
Test Healer
Autonomous test repair in CI & local mode
An agentic workflow built on the Claude Agent SDK that automatically diagnoses and repairs broken Playwright tests — both in CI pipelines and in local development mode. When a test fails, the agent uses Playwright MCP to navigate the live application, observe the current UI state, identify what changed, and apply targeted fixes. This eliminates the manual triage cycle entirely, closing the loop from failure to fix without human intervention.
Mode
Runs in CI pipelines and on-demand locally — adapts behavior to the environment
How it works
Agent navigates the live app via Playwright MCP, observes UI state, diagnoses root cause, and patches the test
Impact
60%+ reduction in manual debugging time. Broken tests are self-healing, not backlogged
Selective Test Execution
AI-driven Playwright test selection for faster CI
Instead of running a 60-minute Playwright suite on every PR, Claude analyzes the pull request diff via the GitHub API, maps each changed file to its impacted test coverage across three tiers — direct component impact, feature-level, and integration-level — and outputs a prioritized, reasoning-backed list of test files. Every selection is explainable via an HTML report uploaded as a Buildkite artifact. A four-condition safety gate and automatic fallback to the full suite ensure CI is never blocked.
50%
Lower CI cost
5–10 min
vs 60 min full suite
10–30×
Compute cost ROI
$0
Rollback cost (env var)
AI Mapper
Documentation-driven test coverage matrix generator
A custom Claude Skill that generates a comprehensive test coverage matrix purely from product knowledge base documentation — no codebase access or code exploration required. It works in three phases: fetches KB articles via a locally-running Rails app's internal API, reads every doc to extract distinct user journeys with step-by-step UI detail and expected results, then cross-references those journeys against existing Playwright specs to classify each as Already Present, Skipped, or Yet to be Added. Output is delivered as a color-coded XLSX, AI-ingestible Markdown, and structured JSON — enabling shift-left test planning before a single line of code is written.
Input
Knowledge base documentation only — no code context, no test file exploration required to get started
Output
Color-coded XLSX, Markdown, and JSON — each journey with priority, steps, expected result, and coverage status
Impact
In one run, surfaced 53 uncovered journeys out of 55 — revealing a 96% coverage gap in minutes, not days
Skills
Core Technical Skills
Agentic AI / LLM Engineering
Cloud, DevOps & Integration
AI Evaluation & Reliability
Client-Facing & Delivery
Languages
Full-Stack / Product Engineering
Experience
Work Timeline
Senior Quality Engineer (AI-Driven Deployment)
Circle
- ▸Deployed directly into Circle Inc.'s engineering org to architect an AI-driven quality platform using Playwright, Claude Code MCP, and Buildkite MCP, powering autonomous, self-healing end-to-end test systems in production
- ▸Designed and shipped Selective Test Execution: a Claude Agent SDK–powered system that reads a PR's git diff, reasons about blast radius, and selects only impacted tests on private branches, with full-suite runs on main as a safety net — cutting CI cost by 50% and speeding up developer feedback loops
- ▸Built intelligent flaky-test recovery and adaptive test-selection strategies that measurably reduced flake rate and CI noise across production suites
- ▸Engineered automated root-cause analysis that posts actionable fix recommendations directly into pull requests, reducing time-to-resolution for engineering teams
- ▸Led org-wide adoption of LLM-powered tooling: authored best practices, playbooks, and enablement programs — the same evangelism and enablement work core to FDE and applied-AI roles
Engineering Manager, Quality
GitLab Inc.
- ▸Led a distributed team of 8+ engineers, architecting CI/CD test infrastructure handling 100K+ executions daily and driving pipeline flakiness from a 15% baseline to a stable low-single-digit range
- ▸Partnered with product and engineering stakeholders to translate ambiguous business requirements into testing strategy — a discovery-to-delivery workflow analogous to FDE client engagements
- ▸Established quality frameworks, SLOs, and OKRs that aligned engineering practices with business objectives
- ▸Drove quality-left strategy by embedding QE practices early in the development cycle, improving deployment frequency
Lead SDET / Senior SDET
Freshworks (Freshdesk)
- ▸Architected test frameworks spanning 20+ microservices and rolled out Consumer-Driven Contract testing org-wide, preventing breaking changes across service boundaries — systems-integration work core to FDE deployments
- ▸Delivered SDK test frameworks enabling external developers to build reliable integrations on the Freshworks platform
- ▸Built scalable CI/CD pipelines using Docker and Jenkins multibranch strategy, reducing deployment time by 50%
- ▸Represented the org at Rootconf, Global Testing Retreat, and VodQA
Staff Quality Engineer
PayPal
- ▸Enhanced end-to-end frameworks for a global payment-processing platform, improving test coverage and reliability
- ▸Codified quality gates and deployment best practices across payment services with cross-functional partners
Senior Software Design Engineer in Test
Chronus
- ▸Designed a web and mobile (Protractor/AngularJS) test automation framework, integrating with Solano CI for rapid feedback loops
- ▸Collaborated with the performance team to identify and resolve bottlenecks using NewRelic
SDET / Application Engineer
Amazon
- ▸Reduced Kindle regression execution from 8 hours to 40 minutes (95%) through intelligent parallelization across 100+ test agents — recognized with multiple awards for technical impact
- ▸Designed and implemented test frameworks supporting Amazon's rapid release cycles while maintaining quality standards
Member Technical Staff
Juniper Networks
- ▸Developed and tested Hostchecker for Juniper's SSL VPN solutions and resolved critical customer escalations through rapid root-cause debugging
Education
Education & Certifications
Postgraduate Program (PGP), Forward Deployed Engineering & Applied AI Solutions
Indian Institute of Technology, Roorkee · FDE Academy (Futurense)
Hands-on program covering agentic system design, LLM orchestration, RAG pipelines, and client discovery — directly informing the applied AI projects above.
AI Evals for Everyone
Certification
Evaluation-driven development for LLM systems: reference-dataset curation, model-eval vs. product-eval design, and building measurable trust in agent behavior.
Community
Leadership & Mentoring
Director, Chennai Chapter
Women Who Code
Led one of India's most active Women Who Code chapters, empowering 2000+ women in technology through educational initiatives, mentorship programs, and community events. Organized 50+ technical meetups and workshops covering topics from test automation to career advancement. Mentored women engineers transitioning into leadership roles and provided a platform for first-time speakers to share their technical expertise. Collaborated with tech companies to create inclusive hiring practices and mother-friendly workplace policies. Featured in multiple publications for advocacy work around building support systems for working mothers in tech.
Speaking
Talks & Presentations
Presented a paradigm shift from reactive debugging to proactive test maintenance at the Women in Tech Global Conference (100,000+ attendees). When tests fail in CI, an AI agent analyzes the failure, navigates the application, identifies the root cause, and proposes fixes—all without human intervention. Covered autonomous test healing using Claude Agent SDK and MCP, GitHub Actions integration, and real-world metrics on reduced MTTR and increased stability.
Presented how to build a fully autonomous test healing pipeline using the Claude Agent SDK that transforms how engineering teams handle E2E test failures. When tests fail in CI, an AI agent analyzes the failure, navigates the application, identifies the root cause, and proposes fixes—all without human intervention.
Comprehensive talk on implementing DevTestOps practices, discussing how to build quality into CI/CD pipelines, automate testing at scale, and foster a culture of continuous quality improvement.
Explored strategies for integrating quality engineering practices into DevOps workflows, focusing on shift-left testing, CI/CD optimization, and building a quality-first engineering culture.
Shared GitLab's approach to excellence in quality engineering, demonstrating how to bake quality into every step of the software development lifecycle through comprehensive testing strategies, automation, and cultural practices.
Advanced DevOps strategies for supercharging delivery pipelines, including automation techniques, infrastructure optimization, and quality gates that enable rapid, reliable deployments.
Demonstrated practical applications of service virtualization with HoverFly to enable independent testing of microservices, reduce external dependencies, and accelerate development cycles.
Presented solutions to 'Integration Hell' in microservices using Consumer-Driven Contracts (CDC) with Pact. Demonstrated how CDC enables asynchronous integration testing as simple as unit tests, eliminating the need to bring up all microservices for integration validation.
Deep dive into implementing CDC testing across microservices architecture, preventing breaking changes, and enabling independent service deployments at Freshworks.
Media
Featured In
Building an Essential Support System
Freshworks Insider
Women in Tech: Importance of Mother-Friendly Workplaces
The News Minute