Y Combinator W26, San Francisco, California
Remove the noise. Keep the signal.
Compresr is a context-compression company. Long prompts drown the answer-bearing tokens in noise: the model reads everything but reasons over a fraction of it. That is context rot: more context, worse answers, higher cost.
We compress context with the query in mind, dropping the tokens your model never uses and keeping the ones that decide the answer. It works horizontally (across providers, models, and the rest of your stack) because the right context matters no matter who generates the tokens.
Why trust the claims
Compression is the founders' research specialty, not a side project. The work is published, the benchmarks are public, and the system runs in production.
- API uptime
- 99.95%
- P95 latency
- ~1.2s
- avg compression
- ~60%
API uptime
P95 latency
avg compression
The founding team
Four EPFL-trained researchers who have spent their careers on efficient ML and context compression.
Ivan Zakazov
CEO
EPFL PhD on LLM context compression. Ex-Microsoft, ex-Philips Research. Started a physics undergrad at 15.
Oussama Gabouj
CTO
EPFL DLab researcher, ex-AXA Research, lead at the EPFL AI team. Efficient ML systems and prompt compression.
Berke Argin
CAIO
EPFL Computer Science, ex-UBS. Top 0.1% nationally in math.
Kamel Charaf
CPO
EPFL Data Science MSc, ex-Bell Labs. Top-20 national scholarship.
Why trust Compresr
EPFL research lineage
Four founders trained at EPFL, and the CEO holds a PhD focused specifically on LLM context compression.
Y Combinator W26
Backed by Y Combinator and building in San Francisco, California and Europe (Switzerland).
See the signal for yourself
Start compressing context in minutes, or talk to the team about an on-prem deployment in your VPC.