SELECTED WORK

Autonomous CRM agents, running in production.

Live Copilot Studio agents that read Salesforce in plain English, orchestrate specialist sub-agents, audit denied claims, and hand governed data to Power BI — each one built inside Microsoft.

01 / CRM AUTOMATION

Ask your CRM anything, in plain English — answered from live Salesforce.

PROBLEM

Getting a straight answer out of Salesforce means clicks, filters, and report builders — one small question, many manual steps.

OUTCOME

A generatively-orchestrated agent chooses the right Salesforce action per question — accounts, contacts, or a raw SOQL query — and returns clean, formatted answers from live data in seconds.

Three Salesforce tools, one agent, zero sign-in: it looks up a company's phone and website, finds a contact's email, or runs an on-the-fly query across any object — deciding which to use on its own.

COPILOT STUDIO GENERATIVE ORCHESTRATION SALESFORCE CONNECTOR SOQL MAKER-SCOPED AUTH

Salesforce CRM Agent — natural-language lookups across accounts, contacts & any field via live SOQL · live agent

Try asking → What's the phone and website for Dickenson plc? List the names and industries of 5 accounts Find the email for a contact

02 / MULTI-AGENT ORCHESTRATION

A multi-agent research syndicate that routes itself.

PROBLEM

Rigid chatbots can't coordinate multiple live data sources or decide what to do next.

OUTCOME

A Generative-AI supervisor delegates each request to specialist child agents, pulls live market and sentiment data, and synthesizes a clean executive report.

A Supervisor agent reads metadata descriptions of its child agents and dynamically delegates to the right specialist — no hardcoded dialogue trees.

COPILOT STUDIO GENERATIVE ORCHESTRATION AGENT FLOWS FINNHUB API NEWSAPI

Financial Research Syndicate — supervisor agent that delegates to specialists and synthesizes a market brief · live agent

Try asking → Give me a market brief on NVIDIA What's the latest sentiment on Tesla?

03 / AGENTIC AI AUDITOR

An agentic auditor that clears denied claims in milliseconds — then writes the appeal.

PROBLEM

Legacy AI billing audits loop through claims one at a time and time out on hundreds of denials.

OUTCOME

A cloud-native Copilot Studio agent runs the audit math in-memory, pinpoints the costliest denial code, and auto-drafts a tailored payer appeal letter.

Flagged denial code CO-45 as 70%+ of lost revenue — and generated a ready-to-send, 3-paragraph appeal in seconds.

COPILOT STUDIO POWER AUTOMATE DATAVERSE IN-MEMORY DATA OPS XPATH/XML

Claims Denial Auditor — audits denied claims and drafts the payer appeal · live agent

Try asking → Audit my denied claims and find the costliest denial code Draft an appeal letter for a CO-45 denial

REFERENCE ARCHITECTURE

An agent that watches your CRM, fixes what's broken, and hands clean data to Power BI.

PROBLEM

Salesforce records drift — duplicates, malformed fields, stale opportunities — and every downstream report inherits the mess.

OUTCOME

A Copilot Studio agent monitors records continuously, resolves anomalies autonomously, and maps governed, clean data into Dataverse for Power BI.

SOURCE

Salesforce CRM (Accounts · Opportunities · Contacts)

1

MONITOR

Copilot Studio agent subscribes to create/update/threshold events.

2

DETECT

Azure AI anomaly checks: duplicate accounts, malformed/null fields, stale records, out-of-policy values.

3

RESOLVE

Governed remediation: merge/dedupe, normalize formats, flag low-confidence cases for human review. Every decision logged.

4

MAP

Clean records written to Microsoft Dataverse: canonical schema, DLP-enforced, audit trail.

ACTIVATE

Power BI reads Dataverse — always-clean dashboards.

CONTINUOUS MONITORING

0 RAW EXPORTS

FULL AUDIT TRAIL