Akash

AI Engineer with 4+ years of experience — 3 in QA engineering, 1 in building and shipping LLM-backed systems to production. The platform engineering background informs how I think about AI in production: deployments need rollback paths, agents need health checks, pipelines need dead-letter queues, and every new LLM call is a latency and cost variable that needs to be tracked. Currently building toward full ownership of agentic AI systems, with hands-on experience in RAG pipelines, agent workflows, and CI-integrated evaluation.

Looking for a team where production discipline is the baseline, not the exception.

AI Engineer (Agentic Systems) at EA Sports (2022-10 – Present)

Building and shipping LLM-backed systems to production with focus on RAG pipelines, agent workflows, and production-grade reliability. Platform engineering background applied to AI systems deployment.

• Built a context-aware documentation assistant for internal developer tooling serving ~800 engineers; indexed 25K+ API docs, changelogs, and runbooks using hybrid retrieval (BM25 + pgvector), reducing average information discovery time from 6 minutes to under 90 seconds.
• Designed a production-grade RAG pipeline using multi-query expansion and query rewriting to improve retrieval recall for ambiguous developer queries, significantly increasing relevant document coverage.
• Implemented advanced retrieval strategies including hybrid search, reranking, and context filtering aligned with multi-stage ('7-layer') RAG architecture, improving answer precision and grounding without increasing latency.
• Applied RAGAS-based evaluation on a 150-question golden dataset; identified chunking issues affecting code context and introduced structure-aware chunking, improving context precision from 0.61 to 0.79.
• Built a LangGraph-based agent for automated incident triage: classified alerts, retrieved runbooks, and generated structured debugging steps; incorporated confidence-based routing to humans, maintaining zero false-positive escalations in the first 45 days.
• Developed a prompt management system with versioned prompts and A/B testing, optimizing response quality, latency, and token efficiency across different use cases.
• Implemented 'sandwich defense' technique to isolate system instructions from user input, significantly reducing risk of prompt injection and jailbreak attacks.
• Added input sanitization and validation layers to filter malicious or malformed queries before LLM processing, improving system robustness.
• Enforced structured output validation using schema-based checks (JSON/Pydantic), ensuring reliable and consistent downstream consumption of LLM responses.
• Built safeguards against prompt injection and jailbreak techniques using layered defenses, retrieval grounding, and strict instruction hierarchy enforcement.
• Designed robust error handling mechanisms across the RAG pipeline, including timeout handling, retries with exponential backoff, and graceful fallbacks to ensure system availability under partial failures.
• Identified and mitigated key failure modes such as stale retrieval after bulk updates, context window overflow, and vector database timeouts under high load.
• Implemented fallback strategies where, in case of retrieval or cache failure, the system degrades gracefully by reducing context size or switching to alternate retrieval paths instead of failing requests.
• Introduced circuit breaker patterns for external dependencies (LLM APIs, vector DB), preventing cascading failures and stabilizing system behavior during outages.
• Built observability into the system with detailed logging and tracing of each stage (retrieval, reranking, generation), enabling rapid root cause analysis during incidents.
• Implemented real-time monitoring of RAG pipelines, tracking latency, retrieval quality, failure rates, and token usage; enabled proactive detection of performance degradation.
• Added token-level cost tracking and budgeting per request class; identified inefficiencies in prompt construction and introduced context truncation, reducing average cost per request to $0.006.
• Optimized latency using Redis-based caching (prompt + response caching), reducing response times by ~40% and lowering overall LLM usage.

QA Engineer at EA Sports (2022-10)

QA engineering responsibilities in AGILE software development environment, testing console gaming applications.

• Responsible for providing Test Cases and executing them within an AGILE software development environment.
• Performing execution of Test Cases, functionality, sanity, and regression testing on Console gaming applications on different platforms.
• Developing the test cases for smoke and functionality tests.
• Validating requirements with test cases and generating reports using JIRA.
• Leading the software quality process involving review of product functions, test specifications & documentation, defect prevention, defect logging & reporting, and resolution of issues.
• Involved in testing on different platforms like Xbox One, PC, PlayStation 4, PlayStation 5, Xbox Series X & Nintendo Switch.
• Requesting build setups from Build Repository Platforms (shift).
• Handling the Stability checks of the game.
• Using Android Visual Studio for taking Crash Logs, TTY files and debug elements for bug posting.

Tester at TATA AIG (2020-11 – 2022-04)

QA testing and quality assurance in insurance domain

• Performing execution of test cases, stability, functionality, sanity, smoke and regression testing.
• Developing the test cases for smoke and functionality tests.
• Validating requirements with test cases and generating reports using JIRA
• Leading the software quality process involving review of product functions, test specifications & documentation, defect prevention, defect logging & reporting, resolution of issue.

B. Tech in Computer Science and Engineering – Veer Surendra Sai University of Technology (2015 – 2018)