Nithish Karanam

Delivered an enterprise conversational AI agent platform for a customer-support team using LangChain, RAG, GPT models, and Pinecone/FAISS. Built ingestion pipelines for PDFs including OCR extraction and recursive chunking to improve retrieval relevance across 50+ internal document repositories. Served agents via REST APIs on AWS Lambda/API Gateway with prompt engineering, caching, and token optimization to reduce inference costs while keeping latency suitable for live chat. Instrumented conversation analytics to classify drop-off, escalation, and unresolved-intent drivers, improving resolution outcomes. Standardized delivery with MLflow prompt versioning and containerized CI/CD, while implementing LLMOps guardrails including RBAC, tool authorization, prompt-injection protection, and PII handling. Integrated agent outputs into CRM and customer-support systems via webhook/API sync.

Led an AI systems integration initiative for an internal e-commerce operations support team by structuring scattered business information into an organized AI-enabled foundation. Connected customer-support databases and CRM records to AI agents via LangGraph, MCP, and RAG for ticket enrichment, knowledge lookup, and case routing, reducing average support resolution time. Built data analysis pipelines using Python, SQL, and Oracle Database 23ai Vector Search to identify churn drivers, escalation patterns, and product issues that informed marketing funnels and retention campaigns. Shipped production-grade FastAPI microservices and REST APIs on OCI, enabling business self-serve access to retrieval, agent orchestration, and data analysis endpoints. Established evaluation and monitoring using LangSmith, DeepEval, and Promptfoo to track retrieval relevance, faithfulness, hallucination rate, and latency.

Designed and maintained scalable ETL/ELT pipelines on GCP using PySpark, Python, SQL, BigQuery, and Dataflow for enterprise reporting workloads. Migrated legacy on-prem IBM DataStage/Db2 batch workflows to BigQuery and Dataflow, improving scalability and operational efficiency. Architected cloud data lakes and analytical models in Google Cloud Storage and BigQuery, building SQL transformations for cleansing, reconciliation, validation, and multi-source integration. Built predictive models for demand forecasting and anomaly detection using scikit-learn, XGBoost, and PySpark MLlib, and supported handoff of scored datasets to business reporting teams. Analyzed funnel drop-off and segment trends to inform marketing spend allocation, improving campaign conversion. Monitored production workflows and implemented data-quality controls, performance tuning, and IAM access governance while partnering with stakeholders to define data requirements and reduce downstream defects.