I Built My Dad a Local AI Translator for His Stock Research

Description and use

Argus Friend Brief turns structured Argus market research into plain‑language, evidence‑linked briefings, generated locally on a laptop using Gemma via Ollama.

Problem

My dad needed a way to digest Argus stock‑research data without navigating complex dashboards.

Outcome

The app runs entirely on a laptop, keeps research data local, and produces evidence‑linked briefings that help the user understand each company’s situation.

Workflow

• Export a read‑only, sanitized snapshot from Argus SQLite
• Run local Gemma 3 4B via Ollama on the snapshot
• Validate JSON output against a Pydantic schema
• Check citations and safety guards
• Display briefing in Streamlit with expandable evidence panels

Evidence

Argus Friend Brief takes a sanitized snapshot of that research and asks a local Gemma model to explain one company in plain language.

Source

The model never sees the database or an open‑ended question. It receives one company's evidence catalog and a Pydantic‑derived JSON schema.

Source

If a response contains an unknown citation or disallowed language, the app gives the model one repair attempt. If the repaired response still fails, the app displays an error instead of an unsupported brief.

Source