Subscribe Work with me
← All work AI agents · Client projectDelivered in 2 days · 5.0 on Upwork2026

Voice-of-customer research engine

Finds how a coach’s buyers describe their problem online, in their own words, with every quote checked word for word.

Voice-of-customer research engine

The problem

A US founder building software for coaches needed his app to learn how each coach’s buyers talk about their problem, from real public comments instead of guesses. I helped him build the research part of his app.

What I built

  1. A research plan from the coach’s intake: Claude names the buyer, the searches to run and the competitors to watch for.
  2. Each source as its own agent: YouTube comments today, Reddit and Instagram (through Apify) switched on by settings, no code change.
  3. Written rules that drop the noise before any AI sees it, then a brief Claude writes with themes, people counts and quotes.
  4. Code that checks every quote word for word against the comment it cites, and five quality checks that say when a brief is too thin. A thin run widens its searches once by itself.
  5. A Vercel API that starts each run on GitHub Actions and writes to Supabase, handed over with a step-by-step setup guide.

In numbers

  1. 2 days

    from contract start to the first live run

  2. 1,461

    public comments read in one sample run

  3. 100%

    of quotes checked word for word against their comment

  4. 5.0

    client review on Upwork

What the client said

“Omer was very quick and very knowledgable. He helped me build a research portion of my app and did it in record time. As I’m building out other tools, he’ll be my first hire to get it done. I’d work with Omer again any day.”

Founder, US coaching-software company · 5.0 on Upwork
Voice-of-customer research engine, screen 1Voice-of-customer research engine, screen 2

Screens show a sample run for a fictional coach, not the client’s data.

Stack

PythonClaude APIYouTube Data APIApifySupabaseVercelGitHub Actions