Voice Agents × System Design Training

Your system design knowledge
isn't the problem.
Your ability to articulate it
under pressure is.

Thinking Gym is a voice AI coach that trains the skill that interviews actually test โ€” structured verbal reasoning in real time. Eight minutes a session. A calibrated score every drill.

Most interview prep builds the wrong skill

๐Ÿ“š

Books & Videos

You recognise a good design when you see one. You cannot produce one under pressure in 45 minutes with someone watching.

Builds: recognition
๐Ÿค

Peer Mock Interviews

Feedback is inconsistent. Your peer is also guessing what the interviewer wants. You leave unsure whether you passed or failed.

Builds: confidence (sometimes)
๐Ÿ’ฌ

ChatGPT

Useful for exploration. Gives no structure, no time pressure, no calibrated score. It agrees with whatever you say.

Builds: familiarity

// All three build recognition. Interviews test production.

What changes when a voice agent is your coach

Traditional prep
  • โœ— Passive โ€” you read, watch, or listen
  • โœ— Self-paced โ€” no time pressure to reveal gaps
  • โœ— Fixed content โ€” same material regardless of your answer
  • โœ— Delayed feedback โ€” days or weeks after an actual interview
  • โœ— No score โ€” no way to measure whether you are improving
Voice agent coach
  • โœ“ Active โ€” you speak, the coach responds to exactly what you said
  • โœ“ Time-constrained โ€” latency is an honest signal of reasoning depth
  • โœ“ Adaptive โ€” follow-up questions change based on your actual answer
  • โœ“ Immediate โ€” concept breakdown within seconds of finishing the drill
  • โœ“ Calibrated โ€” a Thinking Gym Score that moves with your performance
"The engineers who train by speaking their designs aloud โ€” under time pressure, with a coach that responds to exactly what they said โ€” improve faster than those who study passively."

We know voice isn't frictionless. Open-plan offices, commutes, shared apartments โ€” speaking aloud for eight minutes isn't always practical. That's why text mode is a first-class interface. But we believe voice produces better outcomes because it replicates the actual constraint of an interview: no editing, no backspace, no pause.

We're testing that assumption with our first cohort of engineers. Here is what a session looks like.

Two minutes of an eight-minute drill

These are live transcripts from a voice session. The coach speaks, you respond aloud โ€” what you see here is the conversation as it was transcribed in real time.

Design a URL Shortener Step 2 of 6 โ€” Capacity Estimation
2:14
โœ“ Step complete โ€” concepts covered: 3 of 4 DAU estimation  ยท  QPS derivation  ยท  Caching rationale
Design Twitter's Feed System Step 5 of 6 โ€” Scaling Strategy
5:37
โœ“ Step complete โ€” concepts covered: 4 of 4 Fan-out on write  ยท  Hybrid fan-out  ยท  Celebrity account threshold  ยท  Cache invalidation
+38
TGS ↑

Your Thinking Gym Score moves with your performance

After each drill, your TGS updates โ€” up or down โ€” based on which concepts you covered, how many hints you needed, and the difficulty of the drill. It's an ELO-style rating that gives you a single number to improve.

700โ€“900 ยท Beginner 900โ€“1050 ยท Junior 1050โ€“1200 ยท Mid-level 1200โ€“1350 ยท Senior 1350โ€“1500 ยท Staff

We're opening early access to the first 200 engineers.

Founding member pricing โ€” $29/month, rate-locked for life. No credit card required to join the list.

  • โœ“ 1 drill per weekday + 2 drills per weekend day across 8 system design domains (8 minutes each)
  • โœ“ Your Thinking Gym Score โ€” an ELO-style rating that tracks improvement over time
  • โœ“ Full concept breakdown after every session: what you hit, what you missed, what to practice next

No spam. One email when early access opens, one when you're invited.