PRAGYAN P DAS

Domain: Enterprise Customer Support Automation

AgentAssistAI 2.0 – Agentic AI Platform

• Led the evolution to a fully agentic AI architecture, enabling autonomous reasoning, task delegation, and conversational memory.
• Designed and deployed specialized AI agents for contextual summarization of customer history, and real-time decision support.
• Architected a multi-agent conversational system using the Google Agentic AI SDK, leveraging Agent Cards for role-based orchestration and inter-agent communication.
• Implemented tool-calling and agent routing strategies to dynamically handle customer queries and enterprise knowledge searches.
• Built a persistent memory agent using PostgreSQL, enabling long-term conversational context and cross-session intelligence.
• Automated agent deployment to AgentEngine using Azure Pipeline Jobs.
• Established agent observability and traceability using Cloud Trace, enabling performance monitoring, debugging, and governance of autonomous agent behavior.

Tech Stack: Python, Google Agentic AI SDK, Vertex AI, PostgreSQL, Azure Pipelines, AgentEngine, Cloud Trace.

AgentAssistAI 1.0 – GenAI-Powered Assistant.

• Architected an enterprise-scale GenAI chatbot, integrating multi-channel customer interactions (chats, emails, voice calls, CRM cases) into a unified 45-day interaction timeline.
• Designed automated data ingestion and normalization pipelines using BigQuery to process high-volume, heterogeneous customer data.
• Implemented LLM-based contextual summarization to generate concise, actionable interaction overviews for support agents.
• Built a semantic knowledge base search system using Vertex AI Search (RAG architecture) to retrieve and summarize relevant documents from Google Datastore.
• Engineered real-time Speech-to-Text pipelines using Google Speech-to-Text APIs for low-latency call transcription.
• Developed real-time sentiment analysis and post-call AI summarization workflows layered over streaming STT data to identify escalation risks and improve agent efficiency.
• Automated CI/CD pipelines using Azure DevOps for model and service deployment.

Tech Stack: Python, Google GenAI, Vertex AI Search, BigQuery, Google Speech-to-Text, Azure Pipelines, GitHub.

Developed OCR and computer vision pipelines using CNNs, LSTM, TF-IDF, and NLTK for document understanding, and built GCP data pipelines leveraging Cloud Functions, GCS, and BigQuery for scalable processing and storage.

Contributed to the Coastal Management Project, Government of Odisha, performing LULC mapping using LANDSAT imagery, orthophoto generation with DEM, and geodatabase management via GeoServer. Developed GIS models for flood risk and coastal vulnerability analysis, including shoreline change and land erosion assessments, supporting data-driven coastal planning.

Worked on the National Innovation for Climate Resilient Agriculture (NICRA) Project, India, processing field sensor and soil data into spatial datasets using ArcMap 10.3. Performed GIS-based water footprint analysis, applied PCA and Random Forest models to predict crop water requirements, and supported project coordination, reporting, and data quality compliance.

Worked on Tourist Inflow Mapping (DASTA, Thailand) using Python-based vehicle and headcount analysis, ArcGIS point density mapping for hotspot identification, SQL-driven geodatabase management, and continuous spatial data updates, analysis, and map-based reporting.