AI-enhanced for better readability
Syntropic: Founding CTO / GPU Systems Engineer
Company: Syntropic
Location: Remote (US)
Compensation: Equity-heavy
Source: Hacker News
About the Role
We compress LLM KV cache. Our current metrics include:
- 8× compression at 3-bit with 0.43% quality delta (GPU-validated).
- 22× live end-to-end on a 5-tenant workload (quantization × prefix dedup, decomposition always stated), GPU-verified this week.
- 16.89× tenant density on a single V100-32GB.
- Validated across Mistral, Qwen, and Llama with zero re-tuning.
- 90+ patents pending; design-partner pipeline; seed round in motion.
The Challenge
Our current throughput path is a reference implementation. Your first job is the fused Triton/CUDA dequant+attention kernel—turning a validated 22× memory result into production speed. Following that, you will own the serving stack.
Process
- Call
- NDA
- Paid 1–2 week trial on the real kernel
- Mutual decision
We report negative results straight; you should want to work somewhere that does.
Contact
Yosef Elimlich (Founder)
Email: josef@podosai.com
Note: PODOS AI is our affiliated hardware company; the compression IP is Syntropic.