Ved Deo
Verified Expert in Engineering
Data Scientist and Developer
Bengaluru, India
Toptal member since August 26, 2026
Ved is a principal data scientist with over eight years of experience building generative AI and agentic AI systems across the pharmaceutical and telecom industries. His toolkit centers on Google ADK, LangGraph, and Kubernetes. While at Saama Technologies, he reduced API latency by 40% and tripled throughput for clinical trial data standardization platforms.
Portfolio
Experience
- Machine Learning - 9 years
- LangChain - 4 years
- Natural Language Processing (NLP) - 4 years
- LangGraph - 3 years
- AI Agents - 2 years
- Milvus - 2 years
- Google ADK - 2 years
- Model Context Protocol - 1 year
Preferred Environment
GCP, Vertex AI, BigQuery, Cloud Run, Amazon Web Services (AWS), AWS Bedrock AgentCore, Amazon EKS, Docker
The most amazing...
...multi-agent SDTM mapping platform I've architected achieved over 88% accuracy on CDISC clinical trial data standardization for pharmaceutical clients.
Work Experience
Principal Data Scientist
Saama Technologies
- Architected an end-to-end multi-agent SDTM mapping platform on Google ADK, orchestrating six specialized sub-agents (spec discovery, compliance fetch, cross-reference, remediation, report generation, query) with PlanReActPlanner and BigQueryToolset.
- Built a cross-framework agent network connecting Google ADK, LangGraph, and CrewAI agents over the A2A protocol; containerized with Docker and deployed to Kubernetes (EKS) for SDTM dataset generation, validation through the CDISC Rules Engine (CORE).
- Engineered a commercial, pip-installable CLI with a FastAPI back end for remote skill delivery and customer API-key management, integrating Azure OpenAI–powered LangChain Deep Agents and Docker packaging for enterprise deployment.
- Built and deployed dockerized FastMCP servers on Cloud Run, using GCS-backed session persistence to expose CDISC/SDTM tools to Claude, Gemini, and GitHub Copilot via the Model Context Protocol; optimized asynchronous FastAPI handlers for performance.
- Developed a Vertex AI RAG pipeline using Discovery Engine, with stable context pre-extraction and fire-and-poll LRO patterns, to extract all six SDTM Trial Design domains from protocol PDFs using cross-domain cascade logic.
- Architected an earlier RAG-based SDTM mapping pipeline with LangChain, FAISS, and Milvus using MedCPT and OpenAI embeddings, and a multi-agent CSR generation system producing ICH E3 sections and safety narratives from SDTM/ADaM.
- Configured custom MCP servers to extend GitHub Copilot (Chat, Agents, Enterprise) with CDISC/SDTM knowledge bases and clinical validation rules, enabling context-aware code generation across the data science organization.
Senior Data Scientist
Rakuten Symphony
- Developed real-time anomaly detection on MinIO log streams using Keras autoencoders with sliding-window encoding, processing 5+ million daily network events at 95% precision.
- Built a statistical root-cause analysis framework using Granger causality, lag correlation, and p-value analysis to automate incident diagnosis, reducing MTTR by 35%.
- Distributed deep-learning training across multi-node GPU/CPU clusters using TensorFlow, Dask, and Polars, cutting training time by 50%.
Senior Software Engineer (Data Science)
GlobalLogic
- Engineered end-to-end MLOps pipelines on Kubeflow and AI Hub for automated training, versioning, and deployment, supporting 20+ production models and 15+ data science teams with A/B testing and automated retraining.
- Trained and deployed CV models such as VGG16/19, ResNet, InceptionNet, YOLOv3, and EfficientNetB3 for multi-class image classification and object detection.
- Built an autoencoder-based anomaly detection to find out the data points that were showing unusual behavior.
Business Analyst
NuWare
- Delivered time-series forecasting models using ARIMA, SARIMA, and LSTM for bond and equity prediction at a leading U.S. bank, achieving MAPE below 3%; built xlwings-based interactive dashboards that improved sales forecasting accuracy by 15%.
- Architected a time series forecasting model to decide which markets are best for the bond release and what the optimal prices for the bonds should be.
- Built a POC to create a simple pipeline of ML and deploy it on GCP.
Data Analyst
E Progressive Development Services
- Built ML/NLP and predictive modeling solutions across manufacturing and telecom in collaboration with PhD researchers, improving analysis accuracy by 25% and algorithm performance by 30% through advanced feature engineering.
- Architected a statistical model for Huawei to identify the best cluster for customer density and determine which is best for the new tower installation.
- Contributed to two significant projects in this company.
Experience
AI-powered Competitive Intelligence and Lead Generation Platform (SaaS)
Education
Master's Degree in Artificial Intelligence
Liverpool John Moores University - Liverpool, England
Postgraduate Diploma in Machine Learning and Artificial Intelligence
International Institute of Information Technology Bangalore (IIITB) - Bangalore, India
Bachelor's Degree in Electrical and Electronics Engineering
Rajiv Gandhi Proudyogiki Vishwavidyalaya (RGPV) - Bhopal, India
Certifications
Generative AI Specialization
Coursera
Custom and Distributed Training with TensorFlow
Coursera
Skills
Libraries/APIs
Scikit-learn, Keras, TensorFlow, Pydantic, LSTM, PyTorch
Tools
BigQuery, Uvicorn, Amazon EKS, ARIMA, SARIMA, CopilotKit, Apache Airflow
Languages
Python 3
Frameworks
LangGraph, Streamlit, FastMCP, AutoGen
Paradigms
Model Context Protocol (MCP), Anomaly Detection
Platforms
Docker, Langfuse, Cloud Run, Kubernetes, Kubeflow, Vertex AI, Amazon Web Services (AWS), AWS Lambda
Storage
Redis
Other
LangChain, Retrieval-augmented Generation (RAG), Machine Learning, Google ADK, FastAPI, FAISS, Milvus, Prompt Engineering, ChromaDB, Async/Await, WebSockets, GitHub Actions, CI/CD Pipelines, AI Agents, Natural Language Processing (NLP), A2A Protocol, Model Context Protocol, Residual Neural Networks (ResNets), xlwings, GCP, Amazon Bedrock AgentCore, LiteLLM, LLM Fine-tuning, Computer Vision
How to Work with Toptal
Toptal matches you directly with global industry experts from our network in hours—not weeks or months.
Share your needs
Choose your talent
Start your risk-free talent trial
Top talent is in high demand.
Start hiring