Filter this list
Browse all work

All work

Project / active

Capitalyst Brief

A scheduled, source-led pipeline that turns public filings, firm publications, news, and macro data into a cited brief for my own review.

It monitors ten configured research profiles, analyses new material with provenance and entity checks, and archives Markdown and HTML briefs before private delivery by email and Telegram.

The problem

Monitoring the same investment firms each week meant repeatedly checking filings, investor communications, firm publications, and news, then rebuilding the source trail before I could compare what had changed.

I wanted one repeatable research pass without treating generated analysis as finished judgment.

My role

I built and operate Capitalyst Brief for my own research workflow. I set the coverage, source rules, analysis prompts, validation checks, output structure, and delivery cadence.

How it works

  1. Start the run. A weekly schedule—or a manual trigger—starts the pipeline.
  2. Collect the record. Source-specific fetchers gather SEC filings, firm-controlled publications, recent news, and macro context for ten configured research profiles.
  3. Analyse by profile. A model prepares per-profile analysis while the pipeline records provenance, disambiguates entities, and checks certain claims such as assets under management.
  4. Compose with continuity. The prior brief provides context for what changed, while the source appendix is built from collected records rather than model-written citations.
  5. Archive and deliver. The pipeline retains dated Markdown and HTML outputs, then sends the private brief by email and Telegram.

What shipped

The working system is a Python pipeline with scheduled and manual execution, source-specific collection, model-assisted analysis, provenance and validation controls, prior-issue context, deterministic citations, dated archives, HTML rendering, and two private delivery channels.

The retained repository contains multiple dated issues alongside its current Markdown and HTML outputs. That demonstrates repeated operation toward a weekly cadence, not an uninterrupted publication record.

Where judgment stays human

Capitalyst automates collection, drafting, and private delivery; it does not contain a pre-delivery approval gate. I check the source trail, relevance, and conclusions before relying on or reusing the brief. Citation and validation controls make claims easier to inspect, but they do not make generated analysis reliable by default.

Limits

This remains a private, single-user workflow. Coverage depends on external APIs and available public sources, and model output still requires checking. The repository does not substantiate time saved, investment performance, readership, adoption, or commercial demand.

jckng.comThanks for stopping by.

Details and outline
Status
active
Topics
investment-research, automation, applied-ai