SSatnam Singhall work
2025Startup Intelligence · Co-sell

Startup Advisor & Co-sell Intelligence Platform

A multi-agent intelligence platform that unifies startup, engagement, funding, milestone, and co-sell data into account-, portfolio-, and co-sell-level analyses for advisor-facing recommendations.

Azure AI FoundryMicrosoft FabricPowerAppsDataverse
headline impact
Advisor
and co-sell intelligence · account + portfolio + co-sell analysis in one agentic surface
my role
Developer · architecture, agent logic, data integration, Power Platform interaction, and Microsoft Fabric grounding
status
Enterprise pilot · agentic architecture implemented
the problem

What was broken

Startup Advisors lacked consolidated visibility into portfolio health, milestones, funding readiness, engagement, and co-sell opportunities. Workflows depended on manual analysis across fragmented systems, limiting proactive startup activation and consistent co-sell execution.

my approach

How I built it

I built a multi-agent platform on Azure AI Foundry, Microsoft Fabric, Dataverse, Logic Apps / Power Automate, and Power Platform. External data (Founder Hub CRM, Excel) is ingested through Shortcuts, Dataflows, and Pipelines into a Fabric Lakehouse on a Medallion architecture, with DQM checks and a metadata + document index layer. On top, Foundry agents run three layers of analysis — Account-level (startup info, milestone progress, workload usage, growth lever), Portfolio-level (milestone concentration, churn risk, funding events, touch status), and Co-sell (readiness assessment, account status, advisor input, business summary) — feeding a Recommendation Agent that aggregates outputs into advisor-facing actions surfaced through PowerApps with Dataverse-backed telemetry and prompt history.

why this way

The reasoning

Advisor workflows aren't one question — they're three different lenses (account, portfolio, co-sell) on the same startup. Splitting that into dedicated agents under a Recommendation Agent keeps each lens focused and composable. Fabric + DQM gives a single governed data spine; vectorised domain knowledge in AI Search keeps recommendations grounded; Power Platform makes the agent layer reachable from the workflows advisors already use.

architecture

How the pieces fit

01 · Data Sources
Founder Hub CRMExcel files
02 · Ingestion & Data Layer (Fabric)
Shortcut · Dataflow · PipelineLakehouse · MedallionDQM CheckMetadata & Document Index
03 · Agent Layer (Azure AI Foundry)
Account-level AnalysisPortfolio-level AnalysisCo-sell AnalysisRecommendation Agent
04 · Retrieval
Azure AI Search · Vector IndexEmbeddingsSemantic / Vector SearchStructured Startup Data
05 · Frontend & Workflows
PowerAppsPower AutomateSSO · Entra ID
06 · Transactional Store
Dataverse · user dataPrompt history & sample promptsTelemetry
what I built

Key components

  • 01Microsoft Fabric Lakehouse on a Medallion architecture with DQM checks
  • 02Metadata and document index layer feeding the agents
  • 03Account-level agent: startup info, milestone progress, workload usage, growth lever
  • 04Portfolio-level agent: milestone concentration, churn risk, funding events, touch status
  • 05Co-sell agent: co-sell readiness, account status, advisor input, business summary
  • 06Recommendation Agent aggregating outputs across all three lenses
  • 07Azure AI Search vector index over startup domain knowledge
  • 08PowerApps for advisor interaction, telemetry KPIs, and email actions
  • 09Power Automate workflow integration · Dataverse for prompt history and telemetry
  • 010Single Sign-On through Microsoft Entra ID
what I used

Tech stack

Agentic AI
Azure AI FoundryAzure OpenAIMulti-Agent OrchestrationRAG
Data
Microsoft FabricLakehouseMedallion ArchitectureDataflowsPipelinesDQM
Power Platform
PowerAppsPower AutomateDataverse
Retrieval & Security
Azure AI SearchVector SearchEmbeddingsMicrosoft Entra ID
see it run

In-browser demo

A scripted, in-browser walkthrough of a real run, traced step by step. Press play to watch the agents fire.

startup_advisor · weekly portfolio sweep
press play to watch a real run
next up
Automated DQM Frameworkopen