Explore anyone'sdigital footprint.

Traceify loads a public X profile and its recent tweets, then lets you have a grounded AI conversation about the person.

Try@naval@paulg@karpathy

recent tweets read per profile
~100
embeddings searched with pgvector
384-d
cache before a profile is refetched
24h
traceify.anmol0b.xyz/x/naval
The Traceify profile page for @naval: profile stats and top tweets on the left, the AI chat with suggested questions on the right.

From a handle to a grounded answer. Every question is matched against the person's own tweets before a reply is written.

  1. GET /api/profiles/naval

    200 OK

    profilesaved

    tweets100

    retweets, repliesskipped

    cache24h

    01

    Fetch

    The profile and about 100 recent original tweets are pulled from X, with retweets and replies left out, then cached for 24 hours.

  2. bge-small-en-v1.5 · 384 dims

    02

    Embed

    Each tweet becomes a 384-dimension vector and is stored in Neon Postgres with pgvector.

  3. q: “views on startups”

    tweet 140.82
    tweet 030.77
    tweet 570.71

    03

    Retrieve

    Your question is embedded too, and a cosine search over an HNSW index pulls the tweets that match it best.

  4. What do they write about most?

    04

    Answer

    Groq streams the reply token by token, written from the profile, its engagement stats and the matched tweets.

Built with

  • Next.js
  • FastAPI
  • Neon Postgres
  • pgvector
  • fastembed
  • Groq