Commercial case studyResearch prototype

Finding Cultural Signals Hidden Across Unstructured Media

A multimodal knowledge-graph prototype built for Keyllective to explore relationships, narratives, and temporal patterns across approved community media.

See the results
3 problems
  1. 01 Corpus design
  2. 02 Graph build
  3. 03 Exploration
Keyllective
Client
Keyllective
Industry
Hospitality strategy and community placemaking
Implementation
Prototype developed in iterative research sprints
Scope
Hospitality / GraphRAG / Multimodal / Knowledge Graph

01 Original problem

The operational friction before the build.

Before the build, Galang AI found 3 problems at Keyllective.

  1. 01

    Local interviews, podcasts, and video contained valuable cultural context but remained difficult to search or synthesize together.

  2. 02

    Keyword search could retrieve mentions but not reliably expose relationships among people, places, events, and themes over time.

  3. 03

    Manual review was too slow for an exploratory hospitality-strategy workflow.

ThreadFinder GraphRAG linking interviews, places, events, and cultural themes

02 Exact deliverables

What was actually designed and delivered.

Galang AI delivered 4 pieces of work for Keyllective: multimodal ingestion, knowledge graph, graph-enhanced retrieval and conversational exploration.

  1. 01

    Multimodal ingestion

    Transcription and normalization of approved interview, podcast, video, and text sources.

  2. 02

    Knowledge graph

    Entity and relationship extraction designed to preserve links among people, places, themes, and time.

  3. 03

    Graph-enhanced retrieval

    Combined semantic retrieval and graph traversal with source context.

  4. 04

    Conversational exploration

    A lightweight Telegram interface for founder-led research and follow-up questions.

03 Architecture

Multimodal graph-retrieval pipeline

Every answer can be traced back to approved source material while the graph makes cross-source relationships explorable.

  1. 01

    Approved media corpus

  2. 02

    Transcription and normalization

  3. 03

    Entity and relationship extraction

  4. 04

    Knowledge graph plus vector index

  5. 05

    Retrieval and graph traversal

  6. 06

    Source-grounded conversational result

04 Implementation timeline

Prototype developed in iterative research sprints

Exact elapsed dates were not retained in the publication-ready evidence; the page therefore describes the verified build sequence.

  1. 01

    Corpus design

    Select sources, metadata, language handling, and the initial ontology.

  2. 02

    Graph build

    Transcribe, extract entities and relationships, and attach source provenance.

  3. 03

    Exploration

    Add retrieval, conversational access, and reviewer feedback loops.

05 Before and after

Measurements with evidence labels.

Measured facts, observed workflow changes, evaluation targets, and missing baselines are kept distinct.

  1. Source types

    Observed

    Before

    Separate audio, video, and text collections

    After

    Prototype architecture for one queryable multimodal layer

  2. Relationship discovery

    Observed

    Before

    Keyword and manual review

    After

    Prototype graph and semantic exploration with source context

  3. Accuracy benchmark

    Not instrumented

    Before

    No public benchmark retained

    After

    No unsupported precision claim

06 Client voice

Client testimony

“We had hours of interviews, podcasts and local video that nobody could realistically search. ThreadFinder let us ask real questions across all of it and see how people, places and themes connected, with the source right there to check. That is where our best research leads came from.”
Max PalmerCEO, Keyllective

Methodology

How this case was substantiated

  • The case describes the prototype architecture and the research-workflow change Keyllective observed, not a production deployment.
  • No precision, recall, or cultural-validity statistic is claimed without a retained evaluation set.
  • Human domain interpretation remains necessary for cultural conclusions.

Limitations

What the evidence does not prove

  • Output quality depends on corpus completeness, transcription quality, ontology design, and source-language handling.
  • No production deployment or commercial outcome is claimed.
  • Graph connections are research leads, not proof of causation or community consensus.
  • The public case omits private source material and client-specific research findings.

Map the operational friction before choosing the technology.

We can scope the workflow, evidence, adoption plan, security boundaries, and success measurements before recommending a build.

All case studies