Vansh Shah

Bangalore, India

• Own the AI service's event-processing platform — roadmap, design direction, and reliability standards as new event types and consumer services onboard.
• Lead design and code reviews for event-driven services across the accounting and billing stack.
• Mentor engineers on agentic patterns and fault-tolerant service design.

Bangalore, India

• Architected an extensible, fault-tolerant event-routing framework in the AI service with a dequeue-based recovery system — automatic reprocessing of failed events and exactly-once consumption across distributed microservices.
• Owned the Dashboard APIs (Django + PostgreSQL) end-to-end; caching, pagination, and N+1 elimination cut response latency by 60% under production traffic.
• Built agentic workflows with the OpenAI Agent SDK + LangGraph to extract and reconcile bank transactions, with a secondary AI fallback route for non-reconcilable cases.
• Improved credit-card statement extraction accuracy by 70% through prompt and context engineering with Anthropic Claude, deployed in production-scale FastAPI pipelines.
• Designed a high-throughput Tally accounting sync — a Flutter Windows client syncing with backend REST APIs every 10s, moving 500 MB in under 30s for near-real-time books.
• Ran core accounting, billing, and transaction microservices (Django + FastAPI) at under 1% runtime error rate, with GitHub Actions CI/CD and AWS-based monitoring and orchestration (SQS, EventBridge, CloudWatch).

Bangalore, India

• Prototyped agentic workflows and Model Context Protocol (MCP) pipelines as alternatives to traditional GenAI implementations.
• Designed evaluation metrics and testing frameworks to benchmark accuracy, latency, and reliability of LLM-powered solutions.
• Built a production-ready POC classifying invoices, bills, bank statements, and credit-card statements using Microsoft Graph APIs.
• Added unit and integration test coverage across services, uncovering critical edge cases before deployment.