Greater Chennai Corporation Automates Municipal Financial Governance with an Agentic AI Platform on AWS
PUBLIC SECTOR & MUNICIPAL FINANCE · AWS GENERATIVE AI
Project Overview
Greater Chennai Corporation (GCC) is the civic body responsible for municipal governance, public finance, and service delivery across the city of Chennai — administering property tax, trade licenses, E-Way approvals, payroll, and capital projects alongside day-to-day revenue and expenditure operations. Like most large urban local bodies, GCC's financial data was scattered across internal ERP and finance systems, tax and licensing platforms, HR/payroll systems, and project and contract records, supplemented by external feeds from banks, treasuries, and government portals.
GCC engaged ILIOS Digital, an AWS Generative AI Competency Partner specializing in agentic AI/ML platforms for enterprise and public-sector clients, to design a platform that interprets an analyst's question in natural language, autonomously plans and executes the required analysis, and produces an explainable, citation-backed management report — escalating only low-confidence findings for human review. The platform is built on Amazon Bedrock and Amazon Bedrock Agents, grounded in a Bedrock Knowledge Base of GCC's policies and historical reports, and orchestrated across a serverless data and automation layer running securely within GCC's AWS landing zone.
Challenges
Finance analysts, department heads, auditors, and management at GCC routinely needed to answer questions that spanned multiple disconnected systems — identifying revenue gaps or anomalies in tax collections, investigating the root cause of variances, reconciling conflicting figures, and producing management reports that could be trusted and defended. This work depended entirely on analysts manually pulling data from each system, cross-checking it against policy, and compiling reports without a consistent, auditable trail of how conclusions were reached.
Fragmented Data Landscape
Financial and operational data was spread across internal systems (ERP/finance, property tax, E-Way/trade license, HR/payroll, projects and contracts) and external sources (banks, treasuries, government portals, market/CPI indices), making a single trustworthy view of GCC's finances difficult to assemble.
Slow, Manual Investigation
Detecting revenue gaps and anomalies, investigating root causes, and reconciling conflicting information required manual, slow work dependent on individual analysts knowing where the right data and policy documents lived.
Reports Without a Paper Trail
Management reports needed to be explainable and policy-aware; a figure that couldn't be traced back to source data and the applicable circular, SOP, or budget document undermined confidence in the finance function.
Costly Reconciliation
Reconciling conflicting evidence across systems — for example, tax-collection figures that differed between the ERP and a bank feed — consumed disproportionate analyst time relative to the underlying discrepancy.
No Triage Between Routine and Complex Work
Every analytical request, routine or complex, competed for the same limited pool of analyst time and got the same fully manual treatment.
Public-Sector Governance Requirements
As a public-sector body, GCC required strong identity, access, and audit controls, with every data access, agent action, and generated report traceable for governance and compliance review.
Solutions
ILIOS Digital designed an agentic financial-governance platform that interprets a user's question, plans and selects the required analytical path, retrieves the necessary data and policy context, performs anomaly and variance analysis, reconciles conflicting evidence, and generates an explainable management report — escalating only low-confidence findings to a human reviewer. Amazon Bedrock Agents orchestrate this reasoning end-to-end, grounded throughout in a Bedrock Knowledge Base of GCC's policies, circulars, SOPs, budget documents, and historical reports.
Seven-Stage Reasoning Flow
Understand Request
The orchestrator interprets the user's question and underlying intent, submitted via web portal, mobile app, Microsoft Teams, or API.
Plan & Route
The agent chooses the analytical path and identifies which data sources, policies, and prior reports are relevant.
Retrieve & Gather
The agent retrieves the required data, policies, historical context, and precedent from the data and knowledge layer.
Analyze
The agent performs anomaly detection and variance/trend analysis over the retrieved data.
Reconcile
The agent reconciles conflicting data and evidence across source systems, preserving the basis for the reconciled position.
Report & Explain
The agent generates an explainable management report with citations back to source data and policy evidence.
Confidence Check
High-confidence findings are auto-published; low-confidence or exception findings are routed to an analyst or auditor for review before publication.
Platform Architecture
User requests reach the platform through a web portal, mobile app, Microsoft Teams, or API clients, entering the environment through a secure gateway. The Agent Orchestrator, built on Amazon Bedrock Agents, drives the seven-stage reasoning flow and calls Amazon Bedrock for model execution in an isolated inference tier.
The platform reads from and writes to a data and knowledge layer for structured finance data, and a Bedrock Knowledge Base for policy and historical-report retrieval. An events and automation layer handles scheduled events, data transformations, and notifications, while low-confidence or exception findings route to a human-in-the-loop workflow for analyst or auditor review.
The entire environment is secured with identity, encryption, and network controls, and monitored end to end for metrics, audit logging, and tracing.
Key Capabilities
- ✓Natural language question interpretation across web, mobile, Teams, and API interfaces
- ✓Autonomous planning and execution of multi-step financial analysis
- ✓Policy-aware reasoning grounded in GCC's governance frameworks
- ✓Cross-system data retrieval and reconciliation
- ✓Anomaly and variance detection with explainable findings
- ✓Citation-backed reports with full audit trails
- ✓Human-in-the-loop for low-confidence findings
- ✓End-to-end security, compliance, and observability
AWS Services
Agent Orchestration
Services: Amazon Bedrock Agents, Amazon API Gateway
Drives the seven-stage understand → plan → retrieve → analyze → reconcile → report → confidence-check flow across web, mobile, Teams, and API channels.
Foundation Model Inference
Services: Amazon Bedrock
Executes anomaly detection, variance analysis, reconciliation, and report generation in an isolated inference tier.
Policy-Grounded Retrieval (RAG)
Services: Amazon Bedrock Knowledge Bases, Amazon OpenSearch Service
Grounds every analysis in GCC's actual policies, circulars, SOPs, budget documents, and historical reports.
Structured Data & Analytics
Services: Amazon S3, AWS Glue, Amazon Athena, Amazon RDS/Aurora
Provides governed, queryable access to raw, curated, and reported finance data and master records.
Workflow & Event Automation
Services: Amazon EventBridge, AWS Lambda, AWS Step Functions
Coordinates scheduled events, data transformations, and downstream notifications.
Hybrid Connectivity
Services: AWS Direct Connect / VPN
Optional, governed connectivity to GCC's on-premises legacy systems, file shares, and databases.
Identity & Access
Services: AWS IAM, Amazon Cognito (SSO/MFA)
Enforces least-privilege access and multi-factor authentication for every user persona.
Data Protection
Services: AWS KMS, AWS Secrets Manager
Encrypts data and credentials at rest and in transit.
Monitoring & Audit
Services: Amazon CloudWatch, AWS CloudTrail, AWS X-Ray
Provides metrics, anomaly alerting, audit logging, and distributed tracing across the platform.
Resilience
Services: Multi-AZ Architecture, AWS Backup
Delivers high availability and managed backup for disaster recovery.
Return on Investment
GCC's Expected Gains
Time to answer a routine financial-analysis question
BEFORE
Hours to days, via manual multi-system lookup
AFTER (TARGET)
Minutes, via a natural-language request
ROI / QUANTIFIED VALUE
Time reduction: >50% at minimum
Analyst manual-reconciliation effort
BEFORE
100% manual, system by system
AFTER (TARGET)
Reserved for low-confidence exceptions only
ROI / QUANTIFIED VALUE
Manual effort shifted to exceptions
Management report evidentiary trail
BEFORE
Inconsistent, undocumented
AFTER (TARGET)
Citation-backed to source data and policy on every report
ROI / QUANTIFIED VALUE
100% of published reports citation-backed, improving auditability
Share of findings requiring human sign-off
BEFORE
100% of outputs
AFTER (TARGET)
Only low-confidence / exception cases
ROI / QUANTIFIED VALUE
Human sign-off reduced from 100% to exception-only
Anomaly / revenue-gap detection
BEFORE
Ad hoc, dependent on analyst review
AFTER (TARGET)
Systematic, on every relevant request
ROI / QUANTIFIED VALUE
Moves from ad hoc to systematic detection; production detection-rate uplift to be measured
Note: Figures reflect the platform's design targets. Quantified values are derived only where the source provides a measurable before/after baseline; remaining ROI metrics should be validated from production telemetry after go-live.
Strategic Business Impact
- ✓Eliminated analyst bottleneck by automating routine financial analysis and enabling focus on complex, high-value work
- ✓Enhanced public trust through explainable, citation-backed reports with full audit trails
- ✓Accelerated decision-making for municipal leadership with real-time financial insights and systematic anomaly detection
- ✓Improved compliance and governance through complete, traceable audit logs of all data access and agent actions
- ✓Enabled proactive revenue management and variance investigation instead of reactive, ad hoc analysis
- ✓Reduced reconciliation costs by automating cross-system data validation and conflict resolution