• Led end-to-end IVR development as 1 of 2 core engineers on a 10+ member team, owning conversational AI flows for 50K–70K daily calls across multi-region GCP deployments.

• Reduced conversation errors by 40% and stabilized behavior across 270+ regression cases by designing deterministic + LLM-enhanced flows in Dialogflow CX and Playbooks.

• Integrated Gemini-powered reasoning to handle ambiguous user intents, boosting first-call resolution by 25% and cutting repeat call volume across high-traffic telecom flows.

• Cut agent escalations by 30% through structured state management, session-parameter orchestration, and multi-level routing logic.

• Built CI-driven regression and debugging pipelines that reduced QA cycle time by 30% and prevented regressions during the CES 2.0 migration serving millions of end users.

• Drove 35% faster incident resolution by leading cross-functional debugging workflows across QA, product, and platform engineering teams.

• Architected ReX, a personalized RAG-based AI career coach using GPT-4, Vertex AI APIs, Pinecone vector search, and LangChain, achieved 85% retrieval precision@10 on a 120K-document corpus and increased user engagement by 30%.

• Engineered a high-performance LLM inference stack using async batching and Redis semantic caching (75% hit rate), reducing p90 latency by 55% and lowering inference cost by 40%, enabling support for thousands of daily users.

• Developed a production-grade ML pipeline (GitHub Actions, MLflow, Vertex AI) for automated training, evaluation, versioning, A/B testing, and blue-green deployment speeding up release cycles by 75% across 20+ model versions.

• Implemented distributed training using Ray across 4×A100 GPUs for proprietary embedding models trained on 100GB+ datasets, reducing training time from 72h → 28h and enabling sub-200ms retrieval performance at scale.

• Conducted an extensive research review across computer vision, reinforcement learning, and agricultural AI domains to identify gaps in existing approaches (simulation speed, dataset constraints, robustness), shaping the direction of the digital-twin architecture.

• Designed the core digital-twin framework integrating neural surrogate models, computer vision, and reinforcement learning to estimate crop health, fertilizer needs, and environmental stress in real-time, reducing simulation time from hours to <10ms with 95%+ fidelity.

• Trained and optimized ResNet-based CV models and PPO RL agents on 500GB+ multimodal agricultural sensor and imagery data, running over 1,000 controlled experiments to validate architecture choices and improving system accuracy by 25%.

• Contributed foundational CV/RL models and inference pipelines that enabled the product’s eventual deployment across farming networks in India, supporting real-world crop optimization and resource planning.

• Worked on traditional Artificial Intelligence algorithms such as PCA, SVD, NN architectures such as CNN, DNN, RNN.

• Learned about the architecture of Digital Twin and how we can implement digital twin in various engineering domains.

• Studied real-time applications of the digital twin in industries (i.e., Maplesoft, Azure Digital Twin, and so on).

• Implemented some digital twin applications for eight different problem statements.