MCP Protocol · Multi-Agent Orchestration · Vertex AI · Deep Research · Security

Surya Nersu

AI Architect building production autonomous agent platforms with hybrid structured/unstructured deep research, MCP Protocol, and enterprise-grade security.

15+
Years
10+
Agents Built
5K+
Users
60%
Latency Cut
356+
Tests
Download
About

Enterprise AI Leader

15+ years building production autonomous agent platforms for Fortune 500 organizations across energy, market research, healthcare, pharma, and fintech.


Core differentiator: hybrid structured/unstructured deep research orchestration — fusing Vertex AI Search with BigQuery (CHASE-SQL NL2SQL) through intelligent intent routing. Expert in MCP Protocol, Google ADK multi-agent orchestration, Milvus distributed vector search (100M+ scale), and container security (CVE triage, Distroless, SQL injection prevention).

Expertise

Core Competencies

🤖

Agentic AI & MCP Protocol

Production agents with MCP (FastMCP/StreamableHTTP), deep research pipelines, intent routing, multi-agent collaboration.

Google ADKMCP ProtocolFastMCPLangGraphAutoGenReAct
🔎

Hybrid Deep Research & NL2SQL

Vertex AI Search + BigQuery (CHASE-SQL) fusion. 6-route intent routing, relevancy reranking, web grounding with DLP.

Vertex AI SearchBigQueryCHASE-SQLNL2SQLRerankingDLP
🛡

AI Security & Container Hardening

26 CVEs triaged, Distroless Docker, SQL injection prevention, DLP guards, zero-trust IAM, supply chain security.

CheckmarxDistrolessSQL InjectionDLPIAMTwistlock
📊

Vector DB at Enterprise Scale

Milvus distributed (IVF_FLAT/HNSW), sub-50ms @ 100M+ embeddings, gRPC mesh, K8s auto-scaling.

MilvusPineconeFAISSHNSWgRPCKubernetes
🧠

GenAI & LLM Engineering

Gemini/GPT, RAG, PEFT/LoRA/QLoRA, prompt engineering, multi-domain prompts, hallucination detection, guardrails.

Gemini 2.5GPT-4LangChainRAGPEFT/LoRARLHF

Cloud, MLOps & Observability

GCP/Azure/AWS. OpenTelemetry, Cloud Trace, structured logging. CI/CD, Terraform, reproducible builds.

GCP/Vertex AIAzureAWSOTelTerraformDocker
Career

Professional Experience

Aug 2021 — Present
Principal AI Engineer
Microsoft · Redmond, WA
Enterprise Technology Co. · Redmond, WA
  • Architected two production AI agent platforms (Google ADK + Vertex AI) — 10+ agents, MCP Protocol, multi-agent orchestration, 5,000+ users across technology & market research.
  • Designed hybrid structured/unstructured deep research: Vertex AI Search (relevancy reranking) + BigQuery (CHASE-SQL NL2SQL) via 6-route intent routing (incl. HYBRID_COMBINED).
  • Built multi-MCP architecture: Looker MCP (8 tools, FastMCP/StreamableHTTP), Custom MCP (semantic cache), VPC-internal ALB, Private Service Connect, 6-layer security.
  • Developed deep research orchestration: Intent Router → Plan Generator (RESEARCH_TOPIC capture) → Section Researcher → WebEvidenceStage (dual-gate DLP) → Report Composer with unified citations.
  • Led container security: 26 CVEs triaged (Checkmarx/NVD), Distroless Docker (64→9 pkgs), CVE exception framework. Fixed SQL injection in BigQuery ML.GENERATE_EMBEDDING.
  • Achieved 60% latency reduction: CachedMcpToolset, HTTP/2 pooling, parallel tool execution (5→3 LLM calls), shared auth_client.py with thread-safe TokenManager.
  • Engineered Milvus vector DB microservice on K8s: HNSW/IVF indexing, gRPC mesh, sub-50ms @ 100M+ embeddings. Healthcare AI summarization with PHI redaction (87%) & HIPAA compliance.
  • Built feedback & evaluation framework with prompt template versioning, response grading pipelines, and hallucination detection via grounding metadata cross-validation against source documents.
  • Extracted shared core library (-2,000 LOC), 356+ tests (TDD). Multi-domain prompts (6+ verticals). 15+ SQL agents (LangGraph). OTel tracing. CI/CD (GitHub Actions, SonarQube, Twistlock).
Sep 2019 — Aug 2021
Senior Data Scientist
SS&C Technologies · Hyderabad
Enterprise FinTech · Hyderabad
  • BERT/T5 Information Retrieval (90% acc), Doc Recommendation (+83% MAP), PII Redaction (87%), vector search (FAISS/Pinecone), Azure ML/Snowflake/Databricks (+86% perf).
May 2017 — Sep 2019
Lead Data Scientist
Infoshare Systems · Hyderabad
Healthcare AI · Hyderabad
  • AI healthcare chatbot (NLP+EMR), Medical Tracker (TF 91%), RASA.AI, +22% model accuracy, -40% downtime.
Feb 2015 — Apr 2017
Sr AI / Data Expert
MarketLinc · Hyderabad
AI Analytics · Hyderabad
  • Real-time Spark Streaming clickstream for cybersecurity clients. Hadoop/Spark/MongoDB. +15% revenue.
Apr 2010 — Jan 2015
Sr Consultant, Data Analytics
Deloitte · Hyderabad
Big 4 Consulting · Hyderabad
  • ML for clinical trials (Eli Lilly), biostatistics, HIPAA/GDPR, EDC integration.
Sep 2007 — Apr 2010
Software Engineer
Cognizant · Chennai
Global IT · Chennai
  • Java/JEE (Spring, Hibernate) for financial services.
Featured Work

Project Spotlights

Enterprise AI Agent Platform

Enterprise Technology · MCP Protocol · Vertex AI Agent Engine

Architecture

// Multi-MCP Agentic System
UserAgent Engine (Vertex AI / ADK)
  ├── LlmAgent (Gemini 2.5, temp=0)
  │   ├── Looker MCP (8 tools, FastMCP/StreamableHTTP)
  │   ├── Custom MCP (semantic cache search)
  │   └── Local Tools (data trail, headers)
  ├── Conversation Service (Firestore)
  └── Response → User
// Security: 6-layer (network→ingress→IAM→token→MCP→SSL)
// CVE: 26 triaged, Distroless (64→9 packages)
60%
Latency Cut
5K+
Users
6
Security Layers
26
CVEs Triaged
ADK 1.24MCP/FastMCPGemini 2.5Cloud RunOTelDistrolessCheckmarx

Multi-Agent Deep Research Platform

Global Market Research · Hybrid Vertex AI Search + BigQuery · CHASE-SQL

Deep Research Pipeline

// Multi-Agent Deep Research with Business-Driven Routing
User QueryGreeting Agent (welcome, context setup)
  → Intent Classifier Agent (route detection)
  │   ├── BUSINESS_ANALYTICS | DEEP_RESEARCH | HYBRID_INSIGHT
  │   ├── TEXT_ANALYTICS | SOCIAL_LISTENING | CHITCHAT
  │   └── Fallback → escalation handler
  → Research Planner Agent (RESEARCH_TOPIC: model-driven)
  → Section Researcher Agent
  │   ├── Vertex AI Search (EnhancedTool + relevancy reranking)
  │   └── BigQuery (CHASE-SQL: decompose→classify→generate→post)
  → Web Evidence Agent (dual-gate DLP: UI toggle + Remote Config)
  → Grounding Validator (hallucination check via metadata)
  → Report Composer Agent (citations: src-N, web-N)
// Prompt Templates: versioned per-domain (persona, sql, ux_rules)
// Feedback Loop: response grading + prompt refinement pipeline
10+
ADK Agents
6
Route Types
356+
Tests
-2K
LOC Eliminated
6
Agent Types
Google ADKVertex AI SearchBigQueryCHASE-SQLFirebaseFastAPITerraform

Milvus Vector DB Microservice

Enterprise Semantic Search · Kubernetes · gRPC · 100M+ Embeddings

Distributed Milvus on K8s with IVF_FLAT/HNSW indexing, gRPC mesh for sub-50ms retrieval, HPA auto-scaling, health probes. Integrated as RAG backbone for agentic applications.

Milvus 2.xK8s/HelmgRPCHNSWetcdMinIOPrometheus

Healthcare AI Summarization

Clinical Intelligence · PHI Redaction · HIPAA · Medical Imaging

Transformer extractive/abstractive summarization of clinical notes. NER-based PHI redaction (87%). CT tumor segmentation (U-Net). HIPAA-compliant pipelines. Health Bot integration.

TensorFlowPyTorchspaCyU-NetAzure MLHealth Bot
Education

Education & Certifications

NIT Tiruchirappalli

Master's Degree
2005–2007

Andhra University

B.E. — Electrical & Electronics
2001–2005

AWS Solutions Architect

Associate Level
Amazon Web Services

Publications

Cybersecurity ETL · AI Threat Detection · Distributed Data Lakes
Published Research

Explore My Work

67+ repositories: agentic AI, MCP implementations, RAG pipelines, ML artifacts.

View GitHub →
Connect

Get In Touch