Close Read Research
A social listening tool that gives regulated brands an honest, compliance safe read on how they actually show up in Reddit conversations.
I built this because the tools that claim to cover social listening hand back sentiment scores and volume charts, and none of that tells a brand team what people actually believe or what the brand is allowed to do about it. I wanted to find out whether reading the threads properly, one at a time, would produce something a team could put in front of legal and act on.
An end to end pipeline that pulls Reddit conversation data about a brand, classifies and scores it with Claude, and renders a branded, client-facing PDF report. Built for industries where you cannot fire off generic engagement tactics and call it a strategy.
The scoring is deterministic and computed in code rather than left to model narration, which is what makes the same brand produce the same read twice. I ran the same threads through the pipeline repeatedly and documented consistent findings before trusting it on a real brand.
How it works
- Python pipeline: Reddit ingestion, Claude-powered sentiment and severity classification, synthesis, PDF report rendering
- Deterministic scoring engine with a six-bucket recommendation model (fix offline, displace, engage now, acknowledge, amplify, monitor only) computed in code
- Reproducibility tested before use on any real brand, with a documented internal report
- Compliance first: only discloses compliant tactics, never recommends unauthorized in-thread engagement, never names private individuals
Where it stands
- Validated on a live pharma brand's Reddit presence, real threads and real findings
- Shipped with a client-facing methodology page and an internal reproducibility report
- Independently built and owned, separate from employer work