Vikas Kumar Singh
Manager, Agentic AI @ eBay | AI/Backend Architect specializing in Agentic Systems
Bangalore, India
Business Impact at a Glance
Profile
AI Systems Architect & Principal Engineer with 10 years of experience designing and implementing production-grade AI platforms, multi-agent systems, and enterprise data infrastructure across regulated sectors (pharma, retail, healthcare, and financial services). Drives technical execution on high-stakes systems, transforming complex data challenges into scalable, high-performance architectures.
Built and shipped end-to-end production systems in a forward-deployed capacity — serving as the hands-on design authority for low-latency, high-concurrency solutions. Now directing technical architecture and system strategy for Agentic AI at eBay India, continuing that end-to-end engineering ownership model from inside the organisation.
Led technical delivery and mentored engineering teams of 4–28 developers across Fortune 500 engagements. Combines deep technical execution (FastAPI, Databricks, PySpark, custom ML/LLM routing) with architectural leadership, personally authoring core components and defining key systems patterns.
Core Expertise
GenAI & LLM Systems
- Multi-Agent Orchestration (LangChain, LangGraph, MCP)
- RAG Architecture & Retrieval Optimisation
- LLM-as-Judge Evaluation Frameworks
- Token Budgeting, Semantic Caching & Dynamic Model Routing
- Prompt Engineering & Output Validation
System Design & Backend Engineering
- API Design Patterns & Service Architecture
- Service Decomposition & Scalability Trade-offs
- Caching & Queueing Strategy
- OIDC / OAuth2 Authentication Design
- Async FastAPI (concurrent, batching, connection pooling)
Data Platforms
- Databricks Lakehouse (Medallion Architecture)
- Unity Catalog & Delta Lake
- PySpark - billion-row pipelines
- AWS Glue & S3, DuckDB, OLAP design
- Geospatial Feature Management (200M+ features)
ML, MLOps & Engineering
- Bayesian MCMC, XGBoost, SHAP Explainability
- MLflow — model lifecycle, A/B testing, registry
- Docker, Kubernetes
- CI/CD & Automated Compliance Validation
- Regulated Systems: HIPAA / GDPR / Solvency II
Selected Case Studies
Production systems designed under real-world scale, governance, and cost constraints.
Multi-Agent GenAI Analytics Orchestration
Role: Principal AI Architect
Orchestrated the delivery of a production-grade GenAI platform for executive analytics at a large-scale eCommerce client. Engineered a custom async routing layer to bypass framework overhead, automated 60 weekly reports to replace 40 FTE of manual analysis, and slashed LLM inference costs by 65% through semantic caching and dynamic routing.
Read Case Study →Geospatial ML Site Selection & Forecasting
Role: Lead AI Architect
Standardized new-store site selection for a $10B+ retailer by designing a Databricks Lakehouse-based geospatial ML engine. Replaced intuition-driven real estate decisions with a defensible forecasting system that accelerated approval cycles from 3 weeks to 5 days, informing over $50M in monthly capital allocation.
Read Case Study →Professional Experience
eBay India — Manager, Agentic AI
Bangalore, India · Jul 2026 – Present
- Scaling the production GenAI platform to 5,000 users and architecting an operations platform targeted to save 4,000+ man-hours annually (in progress).
- Designed and built backend service architecture (async FastAPI, service decomposition, caching layer) supporting the GenAI analytics platform at 1,000+ concurrent enterprise users at <100ms latency — automating 60 reports/week and eliminating $100K/year operational overhead.
- Implemented SSO authentication (OIDC) for role-based access control across enterprise user tiers.
- Implemented LLM cost-governance layer (token budgeting, semantic caching, dynamic model routing) — cutting inference costs 65% while maintaining output quality.
Tredence Analytics — Data Science Manager & Principal AI Architect
Bangalore, India · Nov 2024 – Jul 2026 · Deployed to large-scale eCommerce client as embedded AI Architect, Aug 2025 – Jul 2026
- Directed and executed a 1TB+ Geospatial Lakehouse migration for Retail Real Estate site-selection — applying ISRO-certified geodata analysis techniques — improving query latency 22× (45s → <2s) and accelerating site-selection approval from 3 weeks to 5 days, informing $50M+ monthly CAPEX decisions.
- Designed Databricks-based data engineering pipelines and GenAI model hosting (RAG-based) for product harmonization across a US retail footwear brand's outlets in 50+ countries — enabling automated cross-region rerouting to resolve out-of-stock issues.
- Decommissioned legacy on-prem ETL and migrated to AWS Glue + Databricks — achieving 30% cloud cost reduction and eliminating $1M+/year system maintenance overhead.
- Enabled the AI Centre of Excellence (CoE) through multiple solution and architecture designs for retail clients — reducing client implementation and planning timelines by 50%.
ZS Associates - Business Technology Solutions Consultant, AI Systems
Pune, India · Apr 2021 – May 2024
- Architected demand forecasting engine for a $10B+ pharmaceutical client: chose Bayesian MCMC over standard econometrics for native uncertainty quantification and faster convergence on irregular clinical patterns. Owned end-to-end: model R&D, Spark feature pipeline (500M+ SKU-day pairs), inference orchestration, MLOps. Economic outcome: $2B supply chain waste prevention over 5 years.
- Implemented Spark-based next-best-action recommendation engine for MSL engagement, interfacing Veeva CRM with MLflow model monitoring and feature drift response.
- Scaled AI delivery organisation: built MLOps platform (MLflow versioning, automated model validation, A/B testing, audit logging) enabling 28 distributed engineers to ship 6 concurrent enterprise projects without workflow bottlenecks. Built compliance automation layer ensuring HIPAA/GDPR validation on every model commit.
Collabera Technologies - Data Engineer
Pune, India · Sep 2020 – Apr 2021
- Architected HIPAA-compliant Real-World Data (RWD) platform: PII detection, field-level encryption, audit logging. Processed 500M+ healthcare records/day while maintaining <10MB per-patient data footprint. Integrated Komodo data for HCP scoring.
- Eliminated BI tool vendor dependency ($2M+ contract cycle): engineered custom Python analytics layer with sub-second response on aggregate queries across 100M+ patient records.
L&T Infotech - Engineer
Pune, India · Sep 2016 – May 2020
- Owned ML-driven insurance risk platform: API-integrated model serving at sub-50ms inference SLA, multi-armed bandit A/B testing framework for 10+ concurrent models, feature store for 500+ features with drift detection.
- Engineered GDPR/Solvency II compliant Spark ETL: immutable transaction logs, cryptographic data lineage, automated compliance validation on every pipeline run.
Certifications
- Generative AI Solutions Architect (2025)
- Databricks Certified GenAI Engineer (2025)
- AI/ML for Geodata Analysis - ISRO
- Master's in Applied Data Science - WorldQuant University
- Certified NLP Developer - Vskills
- Certified Python Developer - Vskills
- Certified Tableau Author - Tableau
- Math for Machine Learning - Amazon
Education
- B.Tech, Electronics Engineering
BVDU College of Engineering, Pune (2016) - Master of Science, Applied Data Science
WorldQuant University (2021) - Diploma, Network Security
BVDU College of Engineering (2015)