Content Discovery Concierge
Outcome
Help a viewer or player find something worth their time right now by combining catalogue metadata, transcripts, entitlements, and viewing history into explainable recommendations that respect rights windows and parental controls.
| Portfolio attribute | Value |
|---|---|
| Primary persona | Viewer or player |
| Primary aggregate | ViewerDiscoverySession |
| Cosmos DB containers | 2 |
| Grounding data | metadata, transcripts, reviews, entitlements, history |
| Development estimate | USD 60-220 monthly dev estimate |
| Runtime options | Python FastAPI and C# ASP.NET Core |
Productivity hypothesis
These measures reflect the workflow, evidence, decision boundary, and industry risk for this use case.
| Impact type | Measure | Conservative internal assumption | Customer measurement |
|---|---|---|---|
| Productivity | Discovery-to-confirmation time | 12-22% reduction | (baseline median discovery-to-confirmation minutes - assisted median minutes) / baseline median minutes * 100 |
| Quality or risk | Accepted recommendations | 8-14 percentage-point improvement | assisted percentage of recommendations accepted - baseline percentage |
| Business outcome | Confirmed playback or purchase conversion | 2-6% relative improvement | (assisted confirmed actions per qualified discovery session - baseline rate) / baseline rate * 100 |
Important: Conservative internal pilot assumptions only; not public benchmarks, commitments, or expected customer results. Validate against a comparable baseline or matched control cohort with mandatory human review.
Rationale: Rights-aware discovery can shorten content selection and modestly improve conversion among comparable qualified sessions.
Pilot guardrail: Verify rights, territory, entitlement, availability, and explicit user confirmation before action.
Customer validation plan
- Baseline the current workflow with representative users and tasks.
- Run the accelerator on the same task mix with human review retained.
- Measure discovery-to-confirmation time and accepted recommendations with the documented formulas.
- Replace internal assumptions with customer-observed results before a business case.
Why Azure Cosmos DB
This AI solution needs an operational data plane, not only a vector index:
- Hybrid vector and full-text retrieval combines semantic recall with exact identifiers, codes, names, and domain language.
- Hierarchical partition keys and metadata predicates keep tenant and domain boundaries in the normal query path.
- Operational JSON, source chunks, workflow and agent state, feedback, and audit records remain together for low-latency reads and governed writes.
- Change feed reacts to source and operational updates without polling.
- Autoscale throughput, configurable consistency, and multi-region distribution support variable demand and enterprise availability.
Domain workflow

| Stage | Actor | Input → output | Approval |
|---|---|---|---|
| 1. Express what to watch or play | Viewer or player | Natural-language mood, time budget, or scene memory → Structured discovery intent | Policy controlled |
| 2. Retrieve candidates across metadata and transcripts | Retrieval agent | Intent vector and taste profile → Ranked candidate titles and scenes | Policy controlled |
| 3. Enforce entitlements and parental controls | Rights and controls service | Candidate titles, region, and profile settings → Playable, age-appropriate shortlist | Policy controlled |
| 4. Explain why each title fits | Discovery concierge | Shortlist with taste and scene evidence → Recommendations with human-readable reasons | Policy controlled |
| 5. Confirm playback or purchase | Viewer | Selected title with any rental or upgrade cost → Playback start or confirmed transaction | Human required |
Technical architecture

Cosmos DB data model
| Container | Hierarchical partition key | Documents | Search paths |
|---|---|---|---|
viewerDiscoverySessions | /tenantId, /viewerProfileId | ViewerDiscoverySession, EntitlementGrant, ViewingHistoryEntry, ProfileControlSetting | /intentVectorEmbedding, /intentVector |
catalogTitleIndex | /tenantId, /titleId | TitleMetadata, TranscriptHighlight | /synopsisVectorEmbedding, /moodVectorEmbedding, /segmentVectorEmbedding, /synopsisVector, /moodVector, /segmentVector |
Query contract
| Query | Container | Scope | Purpose |
|---|---|---|---|
| Profile taste context | viewerDiscoverySessions | single-partition | Read viewing history and control settings for the profile in one partition query. |
| Playable entitlement filter | viewerDiscoverySessions | single-partition | Restrict candidates to titles inside an active rights window for the viewer's region. |
| Scene-level semantic search | catalogTitleIndex | cross-partition-bounded | Knowledge-container query against catalogTitleIndex (bounded fan-out across the /tenantId + /titleId partitions named by the shortlist): find the exact episode or moment a viewer half-remembers, as a bounded fan-out across the title ids returned by the vector shortlist rather than a scan of every viewer partition. |
| Continue watching resume list | viewerDiscoverySessions | single-partition | Surface partially watched titles with a meaningful resume point. |
Python and C# services route each document type to its declared container and hierarchical partition key. ETag-based writes protect mutable aggregates, while append-only records and bounded cross-partition queries preserve scale.
Azure deployment

Clone the source repository, authenticate with Azure, and run:
Set-Location accelerators/media-entertainment-gaming/content-discovery-concierge
az login
./Deploy.ps1 -Runtime python -Location eastus2
Change python to csharp to deploy the .NET implementation.
Explore the implementation
- Launch signed-in demo
- Implementation article
- Domain data model
- Workflow and API contract
- Python application
- C# application
- Domain container Bicep
- Platform Bicep
- Cost and configuration
- Source repository