<p>Established the AI quality practice for the platform's LLM features, defining evaluation criteria, golden datasets, and quantitative release gates for RAG-based assistants and conversational agents.</p><p> Built RAG evaluation pipelines measuring context precision, context recall, answer relevancy, and faithfulness with RAGAS and DeepEval, blocking releases on regression against baseline thresholds.</p><p> Validated Agentic RAG workflows including query rewriting, multi-hop retrieval, tool selection, and retrieval fallback, asserting on agent trajectory and tool-call correctness rather than final answer alone.</p><p> Owned RAG knowledge base quality: ingestion and chunking validation, embedding and vector index integrity, metadata and access-control filtering, and stale-content detection across document refresh cycles.</p><p> Developed autonomous QA agents using LangChain and LangGraph, building a multi-agent Planner-Generator-Healer pipeline that converts user stories into executable Playwright tests and self-heals broken locators, sharply reducing authoring and script-maintenance effort.</p><p> Led AI security and red-team testing of LLM features covering prompt injection, jailbreak, system-prompt extraction, PII and data leakage, and unsafe tool invocation, mapped to the OWASP LLM Top 10.</p><p> Validated input and output guardrails covering toxicity, PII redaction, topic restriction, and response grounding, verifying enforcement while measuring false-positive impact on legitimate traffic.</p><p> Pioneered agentic test case generation and automated defect reporting, cutting manual QA effort and accelerating release cycles.</p><p> Owned end-to-end quality delivery for a military-grade platform, planning and executing quarterly releases across Web, Mobile, and Backend modules.</p><p> Built and maintained end-to-end automation across Web, Mobile (Android & iOS), and Backend using Playwright, Selenium WebDriver, Appium, and Rest Assured, sustaining full regression coverage every release.</p><p> Maintained performance test suites across core modules (Exam Portal, Smart Library, Student Portal) using JMeter and Locust, safeguarding system reliability under load.</p><p> Integrated automation into GitHub Actions with quality gates and Allure reporting, improving defect traceability and stakeholder visibility into test outcomes.</p><p> Partnered in PI Planning to translate feature requirements into test strategy, and designed daily smoke suites to continuously validate backend availability.</p>
