Enterprise AI Solutions Portfolio
Patent-ready agentic automation solutions across banking, healthcare, finance, and operations—delivering measurable business outcomes with audit-grade governance by Ashish Chandra, Partner & Global Head of AI, KPMG.
Explore Projects
Banking Operations
Corporate Actions Autopilot Agent
An agentic automation solution for Corporate Actions processing (dividends, splits, rights) that reconciles announcements, holdings and entitlements, while generating an audit-ready evidence pack.
Business Outcomes & KPIs
60-80%
Reduction in Manual Reconciliation
Automated entitlement reconciliation eliminates manual processing workload
30-50%
Faster Break Resolution
Corporate action break cycles reduced through intelligent automation
~0
Unexplained Deltas
Near-zero unexplained entitlement delta—every output is evidence-linked
Architecture Overview

Data Flow: Issuer Notice PDFs / SWIFT MT564 / ISO20022 → Ingestion & Normalisation → Doc Parser + Structured Extractor → Positions/Holdings + Custodian Statements → Entitlement Rules Engine → Evidence Binder (citations + hashes) → Exception Workbench (HITL) → Entitlement Graph + Audit Dossier Export
Technical Design & Key Components
1
Event Extractor
PDF/text → normalised corporate action event schema
2
Rules Pack Engine
Deterministic entitlement logic with versioning
3
Entitlement Graph
Instrument → event → account → entitlement edges
4
Evidence Binder
Citations + proof hashes + reviewer signature
Governance & Controls
Data Integrity
  • WORM storage for evidence pack (immutable)
  • Full audit trail: timestamp, model version, rule-pack version, reviewer
  • Cryptographic hashing for tamper detection
Access Controls
  • SoD (creator ≠ approver)
  • RBAC via Entra ID / IAM
  • Approval gates with e-signature attestations
Technology Stack
AI Layer
Azure OpenAI / Bedrock / Vertex AI (LLM)
Graph Database
Neo4j / Amazon Neptune
Data Pipeline
Kafka/Event Hub, Spark/Databricks
Security
API Gateway + RBAC via Entra ID / IAM
Demo Walkthrough
01
Upload Corporate Action Notice
System ingests PDF or SWIFT message containing corporate action details
02
Extract Event Details
AI extracts event type, terms, record date, and payment information
03
Compute Entitlements
Rules engine calculates outcomes per account based on holdings
04
Generate Evidence Pack
System shows citations → calculations → delta analysis with full traceability
05
HITL Approval & Export
Human reviewer approves and exports posting instructions with audit trail
Implementation Code Sample
def compute_dividend_entitlement(position_qty, cash_per_share):
    gross = round(position_qty * cash_per_share, 2)
    return {"type": "CASH", "gross": gross}

def build_evidence_pack(entitlement, citations, 
                        model_version, rulepack_version):
    return {
        "entitlement": entitlement,
        "citations": citations,
        "meta": {
            "model_version": model_version,
            "rulepack_version": rulepack_version
        }
    }
Resume Bullets
  • Designed Corporate Actions Autopilot Agent with evidence-linked entitlement graph for dividend/split/rights processing.
  • Implemented deterministic rule-pack engine with HITL approvals and immutable audit trail.
  • Built exception workbench with discrepancy explanations and impact simulation.
Reconciliation
Reconciliation Agent Swarm
Autonomous reconciliation across trade/cash/position breaks using a multi-agent swarm and self-healing matching policies.
Business Impact
40-70%
Reduced Triage Workload
30-50%
Faster Resolution
Continuous improvement: matching policies improve month-on-month through machine learning.
Architecture Flow

Pipeline: Trades + Confirms + Statements + GL → Normaliser → Matching Fabric → Rules matcher + Semantic matcher (embeddings) → Agent swarm for exception reasoning → Match Graph + Break Root Cause + Resolution Playbook
Agent Swarm Design
Matcher Agent
Generates candidate pairs using hybrid deterministic + semantic similarity
Policy Agent
Proposes new matching heuristics based on historical patterns
Safety Gate Agent
Backtests and drift-checks before policy promotion
Governance & Technology
Policy Governance
  • Auto-promotion requires backtest + approval
  • Policy pack versioning + rollback capability
  • Drift detection on break patterns
Tech Stack
  • Vector DB: Pinecone/Weaviate/Milvus
  • Spark + Kafka for data processing
  • MLflow for policy experiments
  • Grafana/Prometheus for monitoring
Demo Workflow
1
Ingest Daily Breaks
System receives unmatched transactions from multiple sources
2
Generate Match Candidates
Matching engine suggests potential pairs with confidence scores
3
Swarm Analysis
Agents propose root cause + recommended action
4
Operator Approval
Human reviews and approves resolution
5
Policy Promotion
New policy promoted after successful backtest
Code Sample: Hybrid Matching
def match_score(rule_score, semantic_score, amount_delta):
    amt_penalty = max(0, 1 - min(amount_delta, 1000)/1000)
    return 0.5*rule_score + 0.35*semantic_score + 0.15*amt_penalty
Resume Bullets
  • Built reconciliation agent swarm enabling continuous policy learning with safe promotion gates.
  • Implemented hybrid matching engine across structured rules + embeddings.
  • Reduced break triage through AI-assisted root cause reasoning.
Client Support
SLA Copilot & Contract Evidence Dossier
Transforms support tickets into contractual SLA compliance dashboards and client-ready evidence packs.
Business Value
50-80%
Reduction in SLA Reporting Effort
Automated evidence collection and compliance tracking
100%
Audit Defensibility
Clause-level traceability for all SLA metrics
Architecture

System Flow: ServiceNow/Jira Tickets + Chat Logs + Calls → Timeline Builder → Contracts (clauses) → Clause Index + Clause Mapper Agent → Evidence Collector Agent → Evidence Pack Generator → Dashboard & Board Report Export
Key Design Components
Clause Ontology
  • Response time requirements
  • Resolution time commitments
  • Exclusions and exceptions
  • Escalation procedures
Evidence Compiler
  • Timestamps with timezone handling
  • Excerpts from tickets and communications
  • Linkage to specific contract clauses
  • Tamper-resistant dossier hashes
Governance & Security
PII Redaction
Automatic removal of personally identifiable information from evidence packs
Contract Access Control
Client-specific RBAC ensuring data segregation
Tamper-Resistant Storage
Immutable storage with cryptographic hashes for audit integrity
Technology Stack
LLM + Retrieval
Elastic / Azure Cognitive Search
Immutable Storage (Object Lock / WORM)
PowerBI/Tableau Export
Demo Walkthrough
Select Client + Month
Choose reporting period and client contract
Generate SLA Report
System generates compliance status across all SLA metrics
Evidence Dossier
Click any SLA metric → evidence dossier automatically produced
Code Sample
export function buildSlaSummary(tickets: any[]) {
    const breaches = tickets.filter(t => t.slaBreached).length;
    return {
        total: tickets.length,
        breaches,
        compliance: 1 - breaches / tickets.length
    };
}
Resume Bullets
  • Built SLA Copilot mapping tickets to contractual clauses with audit-grade evidence packs.
  • Enabled CXO-ready SLA compliance reporting at scale with clause traceability.
Human Resources
Org-Design Digital Twin Agent
A digital twin that simulates org structures and optimises staffing under skills, cost, and utilisation constraints.
Business Outcomes
20-40%
Reduction in Planning Cycles
Faster staffing planning through automated scenario generation
15-25%
Improved Utilisation
Better billability through scenario optimisation
Transparent tradeoff narratives for leadership decisions with explainable recommendations.
Architecture Overview

Data Flow: HR Data (roles/grades/skills) + Demand Forecast → Constraint Optimiser → Scenario Generator Agent → Change Narrative Agent → Org Chart + Staffing Plan + Learning Plan + Risk/Impact Heatmap
Design Components
MILP/CP-SAT Solver
Mathematical optimisation for constraints
Skills Taxonomy
Role-to-skill mapping and gap analysis
Explainable Org Design
Narrates why decisions were made
Governance & Controls
Bias Detection
  • Grade/location neutrality checks
  • Diversity impact analysis
  • Fair allocation algorithms
Policy Compliance
  • Labour policy constraints
  • Approval gates + model risk review
  • Scenario audit trails
Code Sample: Utilisation Penalty
def utilization_penalty(util):
    if util > 1.0:
        return (util-1.0)*10
    if util < 0.6:
        return (0.6-util)*2
    return 0
Resume Bullets
  • Built HR Org Digital Twin for scenario-based org restructuring with constraint solver + agentic explanation.
  • Delivered staffing + redeployment recommendations with governance controls.
Finance
Close-the-Books Agent
Agentic automation for month-end close: variance analysis, JE suggestions, policy checks, audit export.
Business Impact
15-30%
Close Acceleration
Days saved in month-end close process
60%
Reduced Investigation
Variance investigation workload decreased
100%
Audit Readiness
Via causality graph linking JE to source
Architecture & Design

System Flow: GL + Subledgers + Invoices → Variance Analyser Agent → Policies & Controls → Policy Guardrails Engine → JE Proposer Agent → Approvals → Posting & Audit Pack → Causality Graph (JE→Source→Policy→Narrative)
01
JE Template Library
Pre-configured journal entry templates with parameterised policies
02
Variance Decomposition
AI-powered analysis of account variances with root cause identification
03
Control Testing Harness
Automated validation against financial controls and policies
Governance Framework
1
SoD Approvals
Segregation of duties enforced—creator cannot approve own entries
2
Posting Limits
Restricted account protection with threshold controls
3
Full Traceability
Complete audit trail of JE suggestions and approvals
Code Sample: Policy Guardrail
def je_allowed(amount, restricted):
    return abs(amount) <= 5000 and not restricted
Resume Bullets
  • Implemented Finance Close Agent generating JE proposals with policy-driven guardrails and audit-ready exports.
Healthcare
Clinical Prior-Authorisation Agent
Automates prior-auth packets by mapping guideline criteria to patient facts with evidence citations.
Business Value
30-60%
Reduced Manual Preparation
Automated packet assembly saves clinical staff time
40%
Faster Approvals
Fewer denials due to complete, accurate submissions
Strong compliance and traceability with guideline-to-evidence mapping for regulatory requirements.
Architecture

Pipeline: EHR Extract + Labs + Imaging → Patient Fact Assembler → Payer Rules + Clinical Guidelines → Criteria Extractor → Necessity Mapping Engine → Prior Auth Packet Generator → Redaction + PHI Audit Logs
Design Components
Guideline-to-Fact Mapping
Structured mapping of clinical guidelines to patient data
Evidence Anchoring
Document offsets linking facts to source records
Appeal Packet Generator
Automated generation of denial appeal documentation
Governance & Security
HIPAA-Grade Logging
Comprehensive access logging for all PHI interactions
PHI Redaction
Automatic removal of protected health information where required
Clinician Approval
Model output filtering with mandatory clinical review
Code Sample: Criteria Mapping
def meets_criterion(facts, key):
    return key in facts and facts[key] is not None
Resume Bullets
  • Built explainable prior-auth agent generating regulator-grade packets with guideline-to-evidence traceability.
Get Started
These enterprise AI solutions deliver measurable business outcomes whilst maintaining audit-grade governance and compliance. Each solution is designed for production deployment with comprehensive security controls and regulatory adherence.