Silver Skates: An AI-Powered Research Platform for News Sense‑Making

Silver Skates is an exploratory research platform focused on helping me make sense of the modern news environment.

This document describes the platform as it exists today and outlines several research‑driven directions on the roadmap. The project is very much a work‑in‑progress, and my aim here is to emphasize transparency, open questions, and design constraints over polished conclusions.

Motivation

Modern news consumption increasingly rewards volume, outrage, and fragmentation. Readers are often exposed to dozens of isolated headlines without meaningful help in understanding how stories relate to one another, where tensions exist, or what remains uncertain.

The platform I am working on treats news exploration as a research activity, not a feed. Its goal is not to appeal to users predispositions, but to help them see structure: clusters of related reporting, shared themes, disagreements, and gaps.

Current Platform Capabilities

At its current stage, Silver Skates supports an end‑to‑end research pipeline:

1. Article Ingestion: Articles are collected from external news APIs and stored in a structured database using Prisma and Postgres. The ingestion process prioritizes completeness and traceability over aggressive filtering.
2. Embedding and Clustering: Each article is embedded using a lightweight sentence‑embedding model and grouped into clusters based on semantic similarity. Clusters are the primary unit of exploration in the system.
3. Cluster‑Level Summarization: Rather than summarizing articles individually in isolation, the platform generates short narrative summaries at the cluster level. These summaries are constrained to the provided article context and are designed to describe what the reporting collectively covers, where stories align, and where they diverge.
4. Exploration UI: A web interface allows users to browse clusters, expand individual headlines, and progressively reveal more article text. This structure encourages movement from overview to detail, rather than the reverse.
5. Chat UI: A LLM-powered chat interface allows for more freeform discovery of the ingested articles.

Design Principles

Several principles guide the platform’s architecture and UX decisions:

1. Strict context grounding: Language model outputs are constrained to the articles provided. When questions cannot be answered from the available material, the system is expected to say so explicitly.
2. Human‑legible intermediates: Clusters, summaries, and article groupings are treated as first‑class artifacts that users can inspect and reason about.
3. Exploration over authority: The platform avoids framing summaries as definitive conclusions. Ambiguity and uncertainty are surfaced rather than hidden.

What Is Intentionally Missing

Silver Skates is not attempting to:

• Provide breaking‑news alerts or real‑time coverage
• Claim political neutrality or algorithmic objectivity
• Replace investigative journalism

These omissions are deliberate. They reflect my belief that clarity and understanding require different tradeoffs than speed and scale.

If you are interested in exploring the platform as it evolves publicly, please take a look at the repository on GitHub: https://github.com/jeromecovington/silver-skates