ENGINEERING PROJECT
Multi-agent analytical assistant
Connecting natural language and structured data while keeping control of queries and results.
Discuss your projectContext
Context
Business users want to query their data without mastering technical schemas or writing SQL. Yet a naturally phrased question can have several definitions or lead to incorrect queries. The challenge is therefore twofold: simplify access to analysis and control the path from intent to the data consulted and the response.
Digital Place intervention
Digital Place intervention
Digital Place designed orchestration components to interpret and distribute requests, then produce queries against structured data. Development covered a semantic layer, query generation and validation, execution within a controlled access framework, and analytical reporting. A separate evaluation system was also developed to examine the system’s questions, queries, results and failures.
Architecture
Architecture
A business understanding layer prepares requests for orchestration services. Generated queries pass through server-side checks before data access and response preparation.
Technologies : SQL.
Engineering decisions
Engineering decisions
Separate the responding agent from the system that checks it.
This separation makes it possible to test the query chain and its results without confusing response generation with evaluation quality.
Technologies
Technologies
- FastAPI
- LangChain
- LangGraph
- Langfuse
- PostgreSQL
- React
Result & scope
Result & scope
- Analytical assistant developed
A processing chain connecting business questions, structured queries and reporting.
- Separate evaluation
Components for checking queries and responses obtained from test cases.
Explore