Girish Sekar
8+ years of experience across the full STLC and SDLC, spanning manual testing, automation engineering, and Agile delivery, now expanding into AI/ML systems engineering Led a 20-member QA team on a client engagement, implementing a test pyramid strategy that decomposed 3,000+ E2E test cases into unit and integration tests, improving execution speed and long-term maintainability Built test automation frameworks from scratch using Robot Framework (Python), enabling parallel execution across environments and significantly reducing manual regression effort Enhanced existing UI automation frameworks using Java, Selenium Web Driver, BDD Cucumber, Maven, and TestNG Performed extensive manual testing - Black-box, White-box, System, Smoke, Functional, Regression, Re-Testing, and Sanity testing - across multiple enterprise applications and release cycles Executed manual and automated E2E testing for complex payment domain back-office and card issuance workflows, translating business requirements into detailed test scenarios and edge cases Conducted API testing using Postman, REST Assured, Karate, and SoapUI to validate microservices, back-end integrations, and service contracts Performed performance testing using JMeter, validating application response times and behavior under load as part of release readiness sign-off Automated cross-platform mobile testing using Appium across iOS and Android devices Built and maintained UI regression automation using Selenium Web Driver and Playwright (Java Script) for modern web applications Set up and maintained CI/CD pipelines using Jenkins, Azure Dev Ops, and Cloudbees, enabling automated build, test, and regression gating Prepared test plans, test cases, and traceability matrices based on business and functional requirements across analysis, design, and construction phases Collaborated in Agile/Scrum ceremonies, owning test planning, execution, defect triage, and reporting across sprints Independently designed and built Cognition Engine, a local LLM agent with real-time cognitive health monitoring (entropy, repetition, goal-drift metrics) and a 4-level automated intervention system to detect and correct AI output degradation Designed a data quality evaluation framework (ablation studies, quality filter pipelines, coverage-gap analysis) for a research-grade AI data platform, applying systematic experimental methodology to measure and improve dataset quality Ran controlled A/B/C evaluation experiments on LLM behavior and data quality decisions, logging granular metrics to empirically validate interventions - applying test engineering rigor to AI system evaluation Built LVP, a multimodal video preprocessing pipeline, and validated output quality across multiple LLM providers with quantitative semantic similarity testing (93–98%) Seeking to combine strong production QA/automation experience with hands-on AI/ML tooling in an SDET-with-AI or AI/ML Test Engineering role
Experience
Consultant, Testing
Worldline
Test Analyst
Infosys
Business Support Senior
NTT DATA
Software Engineer
ProV Infotech
Software Engineer
CL Infotech
Education
Saveetha University
BE
Jawahar Vidyalaya
12th Grade
Jawahar Vidyalaya
10th Grade
Skills
SeleniumPlaywright (JS)Robot Framework (Python)AppiumTestNG (Java)Cucumber BDDNUnit (C#)PostmanREST AssuredKarateSoapUIJMeterLLM evaluation designmodel output validationactivation steeringhidden-state analysisA/B/C experiment designablation studiesdata quality frameworksMLXPythoneval harnessesClickHousedbtScrapyLangChainFastAPIspaCyCLIPBLIP-2
PROJECTS
Cognition Engine - Self-Monitoring Local LLM Coding Agent
Python, MLX, LLM internals
Built a local coding agent (Apple MLX + Qwen2.5-Coder-7B) instrumented with a custom Cognitive Health Index (CHI) - combining token entropy, n-gram repetition, and hidden-state goal-drift similarity to detect LLM output degradation across multi-turn sessions Designed a 4-level automated intervention system (steer → compact context → escalate) with activation-steering and context compaction to recover degraded sessions without losing task-critical state Ran controlled A/B/C evaluation experiments across intervention strategies, logging 15+ metrics per turn — empirically identifying the exact turn (∼10) where degradation begins
Cleaning Workflow Planner Platform - Research-Grade Data & AI Evaluation Platform
Python, Scrapy, ClickHouse, dbt, LangChain, FastAPI
Designed a formal data quality evaluation framework (data quality, extraction quality, planner quality layers) with defined metrics and gold-standard methodology, plus an ablation study system measuring filter retention/rejection rates and overlap Built a multi-stage quality filter pipeline (perplexity scoring, repetition detection, image duplicate detection, CLIP alignment) and a ClickHouse + dbt data warehouse with staging → dimension → fact layers Built a multi-method enrichment pipeline (rule-based, NER, LLM-based extraction) with confidence scoring, and a LangChain workflow-planning agent grounded in warehouse-backed retrieval to prevent hallucination
LVP (LLM-Ready Video Package) - Bandwidth-Efficient Video Preprocessing for Multimodal LLMs
Python, FFmpeg, Whisper, Multi-provider LLM APIs
Built a pipeline compressing video 50–100× (e.g., 212MB → 1.8MB) while preserving semantic content, using scene-adaptive keyframe extraction and Whisper-based transcription Built provider-agnostic integrations for Claude, GPT-4V, and Gemini, validating output equivalence quantitatively - measured 93–98% semantic similarity across providers Wrote a test suite validating package integrity and end-to-end provider query correctness