Architected and deployed an enterprise Knowledge Intelligence Platform using Retrieval-Augmented Generation

(RAG), indexing 50K+ internal documents to enable natural language querying and automated knowledge retrieval

across business units.

• Designed and implemented end-to-end Generative AI workflows leveraging Large Language Models (LLMs) for

enterprise automation, semantic search, and contextual Q&A systems.

• Built RAG pipelines using LangChain, FAISS/Pinecone vector databases, and HuggingFace embeddings, improving

contextual response accuracy by 25–35% through optimized chunking, retrieval tuning, and metadata filtering

strategies.

• Developed scalable document ingestion and preprocessing pipelines for unstructured data (PDFs, reports, internal

knowledge bases), enabling semantic indexing and structured summarization.

• Reduced hallucination rates by approximately 18% through retrieval re-ranking, prompt engineering refinements,

and structured output formatting with validation checks.

• Deployed containerized AI services via FastAPI and Docker on AWS EC2, handling 5K+ requests/day with sub-

300ms average retrieval latency and high system reliability.

• Fine-tuned transformer-based models for domain-specific NLP tasks using parameter-efficient tuning techniques

and iterative prompt optimization to improve domain alignment and inference performance.

• Implemented monitoring and evaluation frameworks to measure retrieval quality, generation relevance, and system

stability using automated metrics and human feedback loops.

Led development of an enterprise retail demand forecasting system processing 1M+ historical transaction records

to generate SKU-level weekly predictions for inventory optimization.

• Designed and deployed scalable end-to-end ML pipelines from data ingestion and preprocessing to model serving

using Python, scikit-learn, and SQL.

• Built and optimized supervised learning models (classification & regression), improving predictive accuracy by

15–22% through advanced feature engineering and hyperparameter tuning.

• Engineered robust data preprocessing workflows for large structured datasets (1M+ records), improving down-

stream model stability and consistency.

• Productionized ML models via RESTful APIs (FastAPI), enabling real-time and batch inference integration into

client systems.

• Automated model retraining and evaluation workflows, reducing experimentation time by approximately 30% and

improving reproducibility.

• Implemented performance monitoring, validation protocols, and A/B testing to continuously enhance inference

quality and model reliability in production.