Interview – Introduction Speech
Self Intro:
- Hi I am “Subrat Kumar Sahoo”, I am a “Technical Architect” & “Data Scientist”, over 13 years of Enterprise IT Experience.
- Over the last 6 years focused on Data Science, Machine Learning, Generative AI, Agentic AI, and cloud-native AI architecture.
Data Science Experties:
- I started my journey in the Data Science and Applied Machine Learning space, where I worked extensively on solving business problems using data-driven approaches and predictive analytics.
Generative AI Experties:
- On the Generative AI side, I work with frontier models across OpenAI’s GPT-4o and the current GPT-5 series, Anthropic’s Claude — Opus and Sonnet — and Google’s Gemini Pro and Flash, along with open-source alternatives like Llama, Mistral, and Qwen, focusing heavily on prompt engineering, fine-tuning, and building in hallucination guardrails so outputs stay reliable.
Enterprise RAG Experties:
- Enterprise RAG expertise designing high-fidelity retrieval pipelines with Hybrid Vector Search and Semantic Reranking — using Pinecone, ChromaDB, FAISS, and Milvus for vector storage, LangChain and LlamaIndex for orchestration, and Document AI Layout Parser with layout-aware chunking for high-accuracy document ingestion.”
Agentic AI Experties:
- Agentic AI expertise orchestrating autonomous multi-agent systems using LangGraph, Google Agent Dev Kit (ADK), CrewAI, and OpenAI Agents SDK — building production agents with memory, tool integration via MCP, and governance for enterprise-scale deployment.
Cloud Experties :
- On the GCP AI/ML side, I work extensively with Vertex AI — Model Garden for accessing frontier and open models, Vertex AI Embeddings and Vector Search for retrieval, and Vertex AI Pipelines for training orchestration. For document-heavy workloads, I use Document AI for intelligent parsing. On the serving side, I deploy through GKE and Cloud Run, expose everything through API Gateway, and keep it observable in production with Cloud Monitoring.
Big Data Experties :
- For big data, I use Google Cloud’s tools to handle large volumes of information. BigQuery lets me run fast analytics on huge datasets, Dataflow processes data in real time or in batches, and Pub/Sub handles streaming events as they happen. Cloud Composer ties all these pipelines together and keeps them running on schedule, while Cloud Storage acts as the central place where all this data lives before it’s used for analysis or feeding into machine learning models.
MLOPS Experties :
- Built end-to-end MLOps pipelines on GCP — Vertex AI + Kubeflow for training orchestration, MLflow for experiment tracking, and Cloud Build handles the CI/CD pipeline — it packages the code into a Docker container and deploys it to GKE. — serving models via FastAPI on Cloud Run with Cloud Monitoring for production observability.