ravikumargupta-bosch/ai-usecases-showcase

Purpose and scope

A learning repository of runnable LLM-driven manufacturing and finance examples, implemented in parallel in Python and .NET 8. It also contains independent Python tutorial curricula for AI frameworks and patterns; this is a collection of scripts/projects, not a single deployed application.

Primary entry points

Each domain use case is a standalone console program: Python uses main.py, while .NET uses Program.cs within individual projects listed by dotnet/AIUsecasesShowcase.sln. Representative entry points: python/Manufacturing/01-PredictiveMaintenance/main.py, python/Finance/01-FraudDetection/main.py, and dotnet/Manufacturing/01-PredictiveMaintenance/Program.cs.

Use-case execution pipeline

The eight mirrored examples follow the same conceptual flow: load local sample CSV/JSON, convert domain records into an LLM prompt, invoke the selected provider, parse structured AI output, and print a domain result. Domain implementations include maintenance, quality, forecasting, anomaly detection, fraud, news sentiment, loan risk, and portfolio advice. Key .NET stages are visible in dotnet/Finance/01-FraudDetection/PromptBuilder.cs, dotnet/Finance/01-FraudDetection/TransactionRecord.cs, and dotnet/Finance/01-FraudDetection/FraudResult.cs.

Shared AI-provider boundary

Provider selection is centralized behind service abstractions rather than embedded in every .NET example: dotnet/Shared/Core/IAIService.cs, dotnet/Shared/Core/AIServiceFactory.cs, and dotnet/Shared/Core/OpenAIService.cs. The alternative local implementation is dotnet/Shared/Core/OllamaService.cs. Python has analogous shared modules in python/Shared/ai_service.py and python/Shared/json_helper.py. This boundary allows each scenario to target cloud OpenAI/Azure OpenAI or local Ollama/LLaMA through environment configuration.

Data and result boundary

Examples deliberately use file-backed sample inputs rather than databases or production ingestion. Shared loaders separate CSV/JSON decoding from business prompts in dotnet/Shared/Data/CsvLoader.cs and dotnet/Shared/Data/JsonLoader.cs; samples reside alongside scenarios, e.g. dotnet/Manufacturing/04-AnomalyDetection/data/sample_production_data.csv. Typed input/result models in each .NET project establish the expected pre-LLM and post-LLM shapes.

Runtimes and external services

Python requires Python 3.8+ and root requirements.txt; .NET targets .NET 8 via dotnet/AIUsecasesShowcase.sln. Cloud execution uses OpenAI in Python (OPENAI_API_KEY) and Azure OpenAI in .NET (AZURE_OPENAI_ENDPOINT, AZURE_OPENAI_KEY). Local execution depends on an Ollama runtime with a Llama 3 model and AI_PROVIDER=ollama. No persistent infrastructure, web server, or database is indicated.

Tutorial subsystem

tutorials/ is architecturally separate from the industry examples and serves as executable reference material for framework-specific patterns: tutorials/LangChain-LangGraph-LangSmith/, tutorials/AutoGen/, and tutorials/MCP/. Tutorials include basic-to-advanced stages and use cases, covering agent orchestration, RAG, observability, vector stores, local models, and prompt engineering; they have their own dependency and environment boundaries.