Full-stack/2026/Production
Sway — Influence Analytics Platform
Ranks influencers by how central they really are in a network (PageRank, eigenvector, betweenness), groups their audiences with Louvain community detection, then simulates how far a campaign would travel and picks the best people to seed it within a set budget.
Context
Brands tend to overspend on influencer campaigns because reach gets guessed at rather than measured. I built Sway to actually rank influence in a network and estimate how far a campaign would spread before anyone commits a budget.
Approach
- 01Modelled the social graph in Neo4j and computed influence with PageRank, eigenvector, and betweenness centrality via NetworkX.
- 02Segmented audiences using Louvain community detection.
- 03Simulated campaign spread with Independent Cascade and Linear Threshold diffusion models, then solved budget-constrained influence maximization to pick optimal seed sets.
- 04Served it through a FastAPI backend over a dual store — Neo4j for the graph, PostgreSQL for metadata — behind a Next.js dashboard with Recharts.
Notable
- Production-grade modular backend with dependency injection and a full pytest suite.
- Reproducible deployment (Gunicorn + nginx) with documented architecture and API.
- Turns influencer selection into a measurable optimization problem.
Next project
AfricodeLab ERP