Bridging High-Throughput Distributed Systems with Production Agentic AI & Generative Workflows
AI Specialist and Software Engineer with 5+ years of production experience bridging mission-critical, high-throughput distributed payment architectures (PayPal) with applied Generative AI, Agentic workflows, and predictive systems for global automotive operations.
- 🧠 Enterprise AI & MLOps: Architecting end-to-end AI/ML lifecycles — from data pipelines and LoRA parameter-efficient fine-tuning to scalable production deployment on Google Cloud Platform (GCP) and Vertex AI.
- 🤖 Agentic Systems: Deep practical expertise building autonomous multi-step reasoning engines (ReAct, LangGraph, Model Context Protocol / MCP), RAG pipelines with dense bi-encoders, and automated agent evaluation frameworks.
- ⚡ Distributed Engineering: Proven background modernizing legacy architectures into high-concurrency Java 21 (Virtual Threads) and reactive backends handling millions of transactions with strict SLA and 99.99% reliability.
- 📈 Measurable Business Impact: Track record of delivering multi-million dollar annual recurring savings (~$2.6M+), slashing enterprise operational error rates by 43%, and accelerating batch settlement throughput by 80%.
Diensten Tech (Client: Global Automotive Enterprise) | Chennai, India
(Jun 2024 – Present)
- Enterprise AI & MLOps Architecture: Lead end-to-end AI/ML systems supporting global enterprise operations, directing architecture from ingestion and fine-tuning to cloud deployment on GCP and Vertex AI.
- Enterprise RAG Knowledge Hub: Spearheaded HLD/LLD and deployed an enterprise knowledge retrieval engine integrated into Microsoft Teams using FastAPI, Bi-encoder embedding generation, and MongoDB Vector Search / Vertex AI.
- Autonomous Agentic Workflows: Architected autonomous reasoning systems utilizing the ReAct pattern, LangGraph, and Model Context Protocol (MCP) for contextual routing and automated enterprise query resolution.
- LLM Fine-Tuning & Evaluation: Applied LoRA for parameter-efficient domain adaptation on open LLMs; engineered automated benchmarking wrappers integrated with GitHub Actions CI/CD with enterprise Gemini model support via GCP Workload Identity Federation (WIF) and LiteLLM.
- Predictive AI Traffic Gatekeeper: Architected and productionized a predictive ML filter for redundant API detection, unlocking ~$2.6M USD in recurring annual cost savings (Lead Architect, team of 4).
- Predictive Time-Series Modeling: Engineered an LSTM forecasting pipeline with PyTorch, FastAPI, and Tekton on GCP for long-term strategic vehicle and market incentive modeling.
- Agent Governance & Security: Deployed a predictive detection solution analyzing enterprise API gateway and distributed telemetry logs to identify and govern unmanaged agentic workloads.
PayPal | Chennai, India
(Aug 2020 – Jan 2026)
- Deep Learning Sidecars in Production: Designed and deployed a production Keras neural network sidecar predicting settlement acknowledgement errors in VISA/Mastercard response feeds, eliminating intermediate processing hops and increasing transaction success rates by 90%.
- Legacy Modernization to Java 21: Modernized critical legacy settlement dispute batch applications into Java 21 using Virtual Threads, boosting processing throughput by 80% and improving operational fault tolerance.
- NLP & Document Intelligence: Architected NLP/LLM document processing pipelines for automated dispute resolution, reducing manual audit intervention by 36 hours/month.
- Resilient High-Throughput Messaging: Decoupled batch-to-service chains with Kafka MQ and asynchronous worker layers, safeguarding 99.99% availability during peak transaction volumes.
- Telemetry & Analytics: Engineered operational dispute dashboards using React.js and Elasticsearch, reducing root-cause investigation time by 80%.
- Conversion Optimization: Refined checkout wallet workflows and retry heuristics through data-driven A/B testing, lifting checkout conversions by +5.2% and unlocking $420K in monthly GMV.
(Earlier: Back End Developer Intern @ PayPal | Feb 2020 – Jun 2020)
| Project | Highlights & Tech Stack |
|---|---|
| Enterprise RAG Knowledge Hub | Enterprise knowledge assistant with dense bi-encoder vector embeddings, FastAPI, MongoDB vector index, and Vertex AI. |
| Agentic CI/CD & Benchmarking Framework | Custom CLI & GitHub Actions CI/CD pipeline evaluating multi-agent systems, integrating LiteLLM & enterprise Gemini models via GCP WIF. |
| Agentic Reasoning & LoRA Adaptation | Multi-step reasoning agents using LangGraph & MCP. Parameter-efficient fine-tuning (LoRA) validated via local Ollama runtimes. |
| Predictive API Gatekeeper | Transformer-based predictive classifier filtering redundant network payloads, delivering $2.6M/yr in verified cloud compute savings. |
| Settlement Modernization & Sidecar NN | Java 21 Virtual Threads high-throughput settlement pipeline coupled with a real-time Keras error-prediction neural network. |
| Thephonepedia | Full-stack production web application built with Next.js, Django, MySQL, and automated real-time web scraping pipelines. |
- 🥇 Enterprise Leadership Recognition — Commended for architecting the Redundant API Detection AI solution delivering ~$2.6M in annual recurring savings.
- 🎖 PayPal Spot Awards (2021, 2023) — Awarded for technical excellence in ML anomaly detection and critical payment batch optimizations.
- ⭐ HackerRank Problem Solving (Advanced) — Certified (2020).
- B.Tech, Information Technology — Jeppiaar Engineering College (GPA: 8.5 / 10)
- Languages: English (Proficient), Tamil (Proficient), Hindi (Beginner)
📫 Let's Connect & Collaborate!
Feel free to reach out for conversations on Agentic AI, LLM Fine-Tuning, Distributed Architectures, or MLOps.
