Kinage Intelligence

Kinage's analyst team was spending the first hour of every morning manually trawling news sources, newsletters, and social feeds to identify relevant market signals. The process was inconsistent across team members, had no shared taxonomy for what counted as a signal, and meant high-value intelligence routinely surfaced too late. This project replaces that workflow with a structured, real-time intelligence dashboard built around how analysts actually think.

1. 1.
   Signal Ingestion Pipeline: Build a multi-source ingestion system that pulls from news APIs, RSS feeds, and scraped sources into a unified data model on a recurring schedule.
2. 2.
   Author Enrichment: Enrich raw articles with author context through scraping and outreach workflow automation, turning anonymous bylines into scored, trusted sources.
3. 3.
   Analyst-First Interface: Design a filterable dashboard ranked by recency and relevance, built around the analyst's mental model rather than the data's raw structure.
4. 4.
   Signal Taxonomy: Define and codify what counts as a signal, how it should be ranked, and what 'relevant' means to an analyst, before writing a line of UI code.

• ◆
  Workflow Research: Shadowed the analyst team for two mornings to map the existing manual process, identifying where the most time was lost and what decisions analysts were actually making.
• ◆
  Data Modeling: Designed the signal schema, source, author credibility, recency score, topic tag, and analyst relevance weight, before touching the ingestion code.
• ◆
  Pipeline Architecture: Built the ingestion layer with scheduled fetch jobs per source, a normalisation step, and a deduplication pass before anything hits the database.
• ◆
  Deployment & Validation: Deployed to Vercel with environment-based source configuration so analysts could toggle sources without touching code.

The core product decision was what counts as a signal. Not everything in a news feed is relevant, and a system that surfaces everything is no better than the original manual process. I built a tiered taxonomy: Tier 1 signals are direct mentions of portfolio companies or named competitors; Tier 2 are thematic shifts in adjacent markets; Tier 3 is ambient context. Each tier drives different UI treatment and notification behaviour.

Raw articles often have low author credibility signals. The enrichment pipeline scrapes author profiles, cross-references with LinkedIn and X, and scores authors on domain authority and publication track record. High-credibility author scores boost the parent signal's ranking. This turned the dashboard from a raw feed into a trust-weighted intelligence layer.

I defined the signal taxonomy before writing a line of code, what counts as a signal, how it should be ranked, and what 'relevant' means to an analyst vs. a generalist. The data model came from that work. I then owned end-to-end delivery from problem brief to live deployment on Vercel.

Turned a daily manual process into a real-time, structured feed that analysts can filter and act on without leaving one interface.