SportsIQ

WIP

Knowledge platform in progress, combining sports-domain context with multi-agent workflows and graph-based retrieval.

Ongoing · 2026

Role Product direction, backend architecture, knowledge design

Team Solo project

Period Ongoing in 2026

Links Case study only · GitHub private

Multi-agentgraphRAGpgvectorDjangoPostgreSQL

The live deployment is currently offline. This page documents the platform direction and architecture.

Why This Exists

SportsIQ started from movement analysis, but the current direction is broader: a knowledge platform that can store, retrieve, and operationalize sports-domain information with AI assistance. The goal is to connect domain expertise, structured knowledge, and product workflows in one system.

The long-term advantage is not just computer vision, but the ability to turn sports knowledge into usable retrieval, reasoning, and feedback products.

Planned Architecture

→ Multi-agent workflows for retrieval, synthesis, and feedback generation
→ graphRAG to connect entity relations with document-level retrieval
→ pgvector storage for semantic search and knowledge recall
→ Django backend for orchestration, storage, and review interfaces
→ Sports-domain data structures designed for future movement and performance use cases

What I Did

→ Reframed the project from pure motion analysis into a broader knowledge platform direction
→ Designed the retrieval and knowledge layer around multi-agent workflows, graphRAG, and pgvector
→ Defined the backend structure for orchestration, storage, and review interfaces
→ Shaped the domain model so future sports-performance and movement use cases can plug into the same platform

Status

Ongoing in 2026. Architecture is defined and implementation is in progress. Updates will be posted in the blog.

Technical Notes

Platform Direction

The project is moving from pure motion analysis into a broader sports knowledge platform.
Multi-agent workflows are used to orchestrate research, retrieval, synthesis, and structured feedback generation.
graphRAG and pgvector are part of the current retrieval and knowledge-layer direction.

Data Model Snapshot

knowledge_entities
- id
- entity_type
- canonical_name

entity_edges
- source_entity_id
- target_entity_id
- relation_type

documents
- id
- source_type
- summary

document_embeddings
- document_id
- embedding

agent_runs
- id
- task_type
- status