# Estimation platform for an enterprise bank

- Period: 2025
- Domain: Enterprise bank
- Role: Led design and development
- Key result: 80%, less manual estimation effort
- Stack: LangGraph, Azure ML, MLflow, Databricks, PostgreSQL, Entra ID, Jira
- URL: https://farhanbaluch.com/work/estimation-platform

A five-agent supervisor turns unstructured requirements and intake tickets into effort-scored plans, with SME review and an immutable audit trail.

### Problem

Project estimates started from long, unstructured business requirements documents and intake tickets. Estimating was slow, inconsistent between estimators, and hard to audit.

### What I built

- A LangGraph supervisor coordinating research, estimation and two wizard agents, with typed state, rollback and an SME-in-the-loop review step.
- A production regression model (Azure ML, MLflow) as the baseline, with LLM adjustment, confidence intervals and citations.
- A bronze/silver/gold lakehouse, an immutable Postgres audit log, enterprise SSO and two-way Jira sync.

### Outcome

- Manual estimation effort reduced by about 80%.