POC Lab

AI proof of concepts and implementations — from idea to working system.

LiveGenAI Agent

CPC Comparison AI Agent

Multi-Agent Benefits Comparison System

A multi-agent AI system that automates health insurance benefits plan comparisons using Google ADK and Vertex AI Agent Engine — generating color-coded Excel reports with zero manual steps.

Problem: Benefits administrators spent hours manually cross-referencing health plan documents to compare deductibles, copays, and benefit categories — an error-prone, spreadsheet-driven process with no audit trail.

Architecture Highlights

  • Multi-agent orchestration with Google ADK
  • Natural language plan code extraction
  • Automated Excel report with 5 analytical sheets
  • Custom evaluation framework (4 evaluators)
Stack:Google ADKVertex AI Agent EnginePythonHealth Insurance APIsExcel
View case study
LiveRAG System

Enterprise RAG on Snowflake Cortex

Internal knowledge Q&A at enterprise scale

Production RAG system built on Snowflake Cortex Search and Cortex LLM Functions for querying internal documentation, policies, and procedures with enterprise access controls.

Problem: Employees couldn't find relevant information across 10,000+ internal documents.

Architecture Highlights

  • Snowflake Cortex for native vector search
  • Hybrid search (semantic + keyword)
  • Role-based document access control
  • Citation and source attribution
Stack:Snowflake CortexLangChainFastAPINext.jsPython
POCData Platform

AI Data Quality Monitor

Intelligent data validation and remediation

AI-powered data quality system that detects anomalies, classifies issues, and suggests remediation actions using Claude API. Integrated with Informatica IDMC for automated fixes.

Problem: Data quality issues were discovered late in the pipeline, causing costly downstream failures.

Architecture Highlights

  • LLM-powered anomaly classification
  • Informatica IDMC integration for remediation
  • dbt test suite auto-generation
  • Alerting with severity scoring
Stack:Claude APIPythonInformatica IDMCdbtSnowflake
POCGenAI Agent

Evaluator Generator Agent

Auto-scaffold custom evaluators from an agent spec

An agent that takes a Google ADK agent specification (tools, routing rules, output schema) and generates a complete custom evaluator class scaffold — reducing the cost of writing evaluators for every new agent.

Problem: Writing custom evaluators by hand for every new ADK agent is repetitive, slow, and often skipped — leaving agents untested in production.

Architecture Highlights

  • Gemini-powered evaluator code generation
  • Spec-driven: input agent JSON, output Python class
  • Generates EvalTestCase examples automatically
  • Cuts evaluator bootstrapping from hours to minutes
Stack:Google ADKVertex AIGemini 1.5 ProPythonjsonschema
View case study
In ProgressGenAI Agent

Multi-Agent Workflow Orchestrator

Autonomous business process automation

Hierarchical multi-agent system for automating complex business workflows — document processing, data extraction, validation, and downstream system updates with human-in-the-loop checkpoints.

Problem: Manual document processing for onboarding required 3 FTEs and a 4-day turnaround.

Architecture Highlights

  • Supervisor-agent pattern with LangGraph
  • Human-in-the-loop approval gates
  • Document parsing with vision models
  • Complete audit trail and observability
Stack:LangGraphPythonGPT-4MongoDBFastAPIReact

Explore the Code

Some POCs have public repositories on GitHub.

GitHub Profile