2025Agentic Analytics · FabricDemo available

SME&C Agentic Analytics Platform

A custom agentic analytics system that turns natural-language business questions into grounded KPI retrieval, query enrichment, SQL generation, Fabric execution, validation, charts, and business insights.

FastAPIAzure OpenAIMicrosoft FabricRedisSQL

headline impact

Demo

grounded analytics over Microsoft Fabric · KPI, SQL, charts, insights in one response

my role

Role

Lead developer · custom backend, orchestration, agents, FastAPI APIs, SQL generation, validation, Redis history, Fabric SQL integration, and response formatting

status

Status

Demo available · enterprise agentic analytics solution

demo

Demo available

Demo walkthrough available

the problem

What was broken

SME&C stakeholders depended on dashboards and ad hoc reports built by BI teams. Every new business question turned into a new report request or manual analysis across multiple sources, growing a sprawling report repository and pushing real insight generation behind data modelling and integration work. Users wanted natural-language access to KPI, program, and offer information without writing SQL or waiting on a BI cycle.

my approach

How I built it

I built a custom agentic analytics service that lets users ask natural-language questions over Programs and Offers. A FastAPI endpoint accepts the query, sample row count, and username, starts a Redis-backed history check, warms connections asynchronously, and runs an orchestrator pipeline: intent classification, query exploder, KPI planning, KPI / program metadata retrieval through Azure AI Search, query enrichment, table extraction, SQL code generation, execution against the Fabric SQL Endpoint, a validation / refinement loop, then chart generation and business insight narration. The response is a structured multi-tab payload containing output, insights, follow-ups, HTML table, base64 graph, SQL, metrics used, filters, and confidence — designed for multi-KPI answers with parallel processing.

why this way

The reasoning

One giant text-to-SQL prompt collapses on a real enterprise KPI catalog. Splitting reasoning into intent → KPI planning → enrichment → SQL → validation → insight gives each step a focused contract, evals, and retry boundary. Grounding through Azure AI Search over KPI and program metadata keeps SQL honest; the Fabric SQL Endpoint keeps execution on governed data; Redis keeps user history and warm connections fast; parallel execution keeps multi-KPI answers responsive.

architecture

How the pieces fit

Client / API

FastAPI analytics endpointsRequest: user_query, sample_rows, username

Session & Warmup

Redis history & metadataAsync connection warmup

Orchestration

Intent ClassifierQuery ExploderQuery Enrichment

Knowledge Retrieval (Azure AI Search)

KPI PlannerKPI MetadataProgram MetadataTable & Column Retrieval

Code & Execution

SQL Code GenerationFabric SQL Endpoint ExecutionValidation & Refinement Loop

Response Synthesis

Chart Recommendation & GenerationBusiness Insight AgentStructured Multi-Tab Response

what I built

Key components

01FastAPI analytics API with Pydantic request and response models
02Redis-backed user history, conversation metadata, and connection warmup
03Azure OpenAI agents across intent, KPI planning, enrichment, SQL, validation, and insight
04Azure AI Search retrieval over KPI, program, and table & column metadata
05Microsoft Fabric SQL Endpoint for governed query execution
06SQL validation and retry / refinement loop for resilient generation
07Parallel execution for KPI / program retrieval and insight / chart generation
08HTML table + base64 graph + structured tabs for output, insights, follow-up, SQL, metrics, filters, confidence, and status
09Configuration-driven model deployment, token settings, index names, retrieval parameters, and SQL execution limits
10Observability via logging and per-stage timing breakdowns through Azure Monitor

what I used

Tech stack

Backend

FastAPIPythonAsyncIOPydanticREST APIs

Agentic AI

Azure OpenAILangGraphMulti-Agent OrchestrationPrompt EngineeringRAG

Data

Microsoft FabricFabric SQL EndpointPySparkSQLDelta Lake

State & Ops

Azure RedisAzure AI SearchAzure Monitor

video demo

Supabase-hosted walkthrough

SME&C agentic solution demo

see it run

In-browser demo

A scripted walkthrough of the demo flow, traced step by step.

smec_platform · /analyze

press play to watch the trace

next up

Startup Application Review Agentopen →