research · Shipped — running in production

Auto-research engine

Karpathy-pattern autoresearcher. Weekly competitive intel + content briefs per brand.

What it does

Runs a perpetual research loop on a topic or competitor. Pulls public sources, deduplicates, summarizes, and produces a weekly intel brief plus a list of content angles ranked by opportunity.

Inputs

Topic / competitor list / brand context document. Refresh cadence (weekly/biweekly/monthly).

Outputs

Weekly Markdown intel brief delivered via Slack or email. Optional Notion/Airtable sync for structured tracking.

Cost per run

~$2/run with default depth (Claude Sonnet + targeted web fetches).

Models used

Claude Sonnet for synthesis. GPT-4o-mini for first-pass deduplication. Gemini Flash for fast fact-checking pass.

Best for

B2B brands competing in noisy spaces. Founders who used to spend Friday afternoons reading competitor blogs.

Not for

Brands operating in fast-moving regulated industries where weekly cadence misses critical updates — use a daily skill variant instead.

Common questions

How is this different from running ChatGPT manually?

It runs without you, on a schedule, with deduplication and prioritization tuned over months of internal use across our own portfolio. The output format is consistent week-over-week so you can spot trends.

Can I tune what 'opportunity' means?

Yes — opportunity is scored by a configurable weight vector (search volume, brand fit, competitor coverage gap, freshness). Default weights are tuned for B2B agency / SaaS.