Equity Research Desk

A boutique family office replaced a Bloomberg trial, six spreadsheets, and a Slack channel of links with one screen. Quant scoring, filing NLP, and sentiment in one dataset.

Client
Boutique family office, sub 500m AUM
Industry
Wealth management and family offices
Shipped
2026
01 · Problem

What they were running on

Three analysts, a Bloomberg trial they could not justify post conversion, and a master spreadsheet that broke every time the SEC changed a tag. Filings read on PDF. Reddit copied into a Slack channel. Recommendations debated in a Monday standup, written up in a Word doc, lost on a shared drive. Most decisions made on conviction, not on the file.

02 · Solution

What we shipped

One in-house research platform. Yahoo and SEC pipelines hydrated nightly. Piotroski, Altman Z, and Beneish M scored on every covered ticker. Reddit and X scraped for sentiment with signal tagging. An AI review agent reads each new 10-K and 10-Q against the prior filing and flags tone shifts, removed disclosures, and new risk factors, each tied to the page it came from. The system writes a draft thesis. The analyst confirms or kills it. Recommendations versioned, audit trailed, exportable for IC.

03 · Outcome

What changed after

The Bloomberg debate ended. Coverage doubled with the same headcount. Filings get read the day they post, not the week the analyst gets to them. The IC opens one screen on Monday. The drafts are on it. They argue the call, not the data.

247Tickers under coverage
3Scoring models run on every ticker
Same dayFilings reviewed the day they post
Features

What it does

  • Multi source data pipeline

    Yahoo, SEC, Reddit, X. Raw responses stored as JSONB and never overwritten. Calculated metrics live on top, fully reproducible from source.

  • Piotroski, Altman, Beneish

    Three battle tested scoring models with individual components stored for transparency. Composite 0 to 100 weights quality, value, growth, momentum, and safety.

  • Filing review agent

    Each new 10-K and 10-Q runs against the prior. Tone shifts, removed disclosures, and new risk factors are highlighted in the filing itself, cited by page, and tone is scored section by section against last year.

  • Sentiment signal detection

    Reddit and X mention volume plus per post sentiment. PANIC, EUPHORIA, OVERLOOKED, and TURNING signals fired when patterns cross thresholds.

  • Audit trailed recommendations

    Every BUY, HOLD, and SELL is versioned and linked to the inputs that produced it. IC compliant, reproducible, exportable.

Tech stack
  • Next.js 16
  • React 19
  • TypeScript
  • FastAPI
  • Postgres + JSONB
  • Celery + Redis
  • LLM review agents