In progress 1
Meridian
An AI research analyst that interviews HR software buyers by voice or chat and matches them to verified vendors.
232
commits
Jake Paul
Founder @ VibeWorks
3
shipped projects
158
conversations scored
Conversations read per week
21 Jun to 6 SepBeyond a plain chat
AI fluency
Read from 158 conversations
Novice
Functional
Proficient
Advanced
Advanced
Native
Operating at an Advanced level · step 4 of 5 · 1.7 points from Native
Multi-session arcs, strong prompt construction and adversarial checking on consequential work.
6.8
/10
How Jake works with AI
Founder @ VibeWorks
Jake sits in the Advanced tier on the strength of sustained, high-volume use with genuinely deliberate practice: context is loaded upfront, complex problems are decomposed into sequenced sub-asks, and most work moves through multi-turn refinement arcs rather than one-shot requests. They score consistently in the seven range on description, delivery, and diligence, with discernment close behind — pushing back on substance, soliciting dissent, and occasionally tracing claims to their primary source. What holds the profile below the top band is delegation at 6.0: Jake routes work out to external builders and connected tools competently, but the loop back — bringing results and errors home to diagnose and re-direct — is not yet reliable.
Professional background
Since 2016, he has worked in recruiting, talent, and product roles. His roles progressed from Head of Talent and Recruitment Consultant to Co-Founder + CEO, then Co-Founder and Chief Product Officer, Strategic Advisor, and Product + Innovation; he studied psychology.
Kyle & Co Crosschq TalentWall™ by Crosschq
University of Missouri-Saint Louis Miami University
158 conversations scored
3 projects, 2 verified from GitHub
AI fluency · five dimensions
Read from the conversations Jake really had with an AI assistant, every one of them, not a quiz or a self-assessment. Click a dimension to see what it measured and what the read found.
Read from 156 conversations
6.8 /10
Operating at an Advanced level
Read from 156 conversations, 15 June to 8 September 2026. Strongest on Description, with Delegation the dimension to grow.
Standout strength · Description
Top of five, 7.3
Jake consistently front-loads context, names the deliverable and its audience, and breaks multi-part work into sequenced sub-asks rather than issuing one vague request. They set constraints and guardrails up front so the model can execute in a single pass, and they assign roles when framing matters.
Where to grow · Delegation
Lowest of five, 6.0
Jake's biggest growth opportunity is delegation. The pattern is visible in the strong sessions as much as the weak ones: they compile critique into a build spec for a downstream coding agent, or generate a handoff prompt for a separate tool, and the thread ends at the spec. When a return leg does appear, it can degrade into rewriting prompt wording as a proxy for debugging — as in the mobile quiz session where nothing was confirmed working despite several rounds of increasingly forceful instructions, because the actual coding-agent output was never brought back. Moving up a level means closing the loop by default: after issuing a spec, paste the executor's real result or error back into the same thread, diagnose against the original brief, and re-issue a corrected instruction rather than a rephrased one.
Dimension 1 · Prompt construction and quality
Description
7.3/10
What this measures
How clearly they specify what they want: goals, constraints and examples.
Jake consistently front-loads context, names the deliverable and its audience, and breaks multi-part work into sequenced sub-asks rather than issuing one vague request. They set constraints and guardrails up front so the model can execute in a single pass, and they assign roles when framing matters.
Score over time
Scored once so far. A second read draws the line.
Projects
3 shipped projects.
Selected work in Jake's order, most important first. Each card includes a short note on why they built it.
-
1
Verified Verified from GitHub · 26 of 26 commits are theirs -
2 In progress
Groveros
One-tap personal capture system: press the iPhone Action Button, speak, and the thought lands classified in a Today dashboard.
Why I built it
Built so that nothing captured in a moment gets lost, and so that follow-through is prepared for Jake instead of left to memory, with ADHD-friendly, no-shame design throughout.
Verified Verified from GitHub · 118 of 118 commits are theirs, 41 made with AI145 commits -
3 
In progress
PenCast
PenCast helps newsletter creators turn written posts into podcast-ready audio for public or paid subscriber feeds.
Why I built it
I made this to help Substack creators turn written content into audio, so readers can listen and creators do not need to become traditional podcasters.
From an uploaded archive · unverified143 commits
Tools and platforms
Everything Jake builds and thinks with, grouped by job. A pink mark means it was found in a repository or archive; a dashed mark means it was listed on a project.
Assistants
self-reported
- Anthropic 2
- OpenAI 2
Coding agents
all verified
- Verified Claude Code 2
Build platforms
all verified
- Verified Replit 3
Source and delivery
self-reported
- Cloudflare Workers
Data and docs
self-reported
- Postgres 2
- PostgreSQL
Services
self-reported
- Resend 2
- Stripe
- Twilio
Other tools
self-reported
- ElevenLabs 3
- Alembic
- Buzzsprout
- Claude API
- Cloudflare D1
- Cloudflare R2
- Deepgram
- Drizzle ORM
- FastAPI
- Google Calendar
- Google Gmail
- LiveKit
- Passport.js
- React
- Render
- Replit Object Storage
- Vite
Verified
Verified through a project
Self-reported on a project
Counts are projects where the tool appeared
How Jake uses AI
156 conversations over 12 weeks, about 12.7 a week. 27% ended with work shipped, and 93% of the work that could be checked was checked before it was used.
What for
-
Strategy and planning 37%
-
Building software 19%
-
Writing 18%
-
Research 9%
-
Analysis and data 8%
How it ends
-
Still working it 59%
-
Shipped 27%
-
No task in sight 13%
-
Set aside 1%
Checking the work
-
Tested 73%
-
Stress tested 19%
-
A quick look 7%