Managing Codex-Spark: 1000 Tokens Per Second Changes Everything
I spent a week with GPT-5.3-Codex-Spark on Cerebras hardware. The speed broke every workflow I had.
0xSero
Building the future of AI
I work to make AI more accessible, affordable, and useful for everyone. From compressing LLMs to fit on edge devices, to educating people on the technology and what can be done with it.
60k+
Total Stars
200+
Public Repositories
5k+
Stars Contributed
Open Source Work
Tools and libraries built in public.
TurboQuant
Near-optimal KV cache quantization for LLM inference (3-bit keys, 2-bit values) with Triton kernels + vLLM integration
Python1.2k stars
Parchi
Your AI friend right in your browser. Browser extension for AI automation with orchestrator mode and vision
TypeScript525 stars
vLLM Studio
Control panel for VLLM, SGLang, llama.cpp, ExLlamaV3. Local inference management UI
TypeScript502 stars
MoE Compress
Model-agnostic MoE compression automation: calibration, REAP/quantization, benchmark, publish
Python177 stars
Pi Brain
Local AI memory system with graph-based retrieval and MCP server support
TypeScript162 stars
REAP Expert Swap
How many experts do we need to serve a model? MoE expert pruning research
Python122 stars
SELECTED CAPABILITIES
I take on complex projects at the edge of what's possible and make them boringly reliable.
AI / AGENTS
LLM infrastructure, multi-agent orchestration, tool-calling systems
DEV TOOLS
Open-source libraries, browser automation, data extraction tooling
PRODUCT / SYSTEMS
Interfaces that match the depth of the underlying system
Recent Work
What I've been building lately
Open Source
TurboQuant
KV cache quantization for LLM inference
Open Source
vLLM Studio
Control panel for local inference
Open Source
Parchi
AI friend in your browser
Open Source
Pi Brain
Local AI memory system
Contributed To
Ethereum FoundationZKSyncGitCoinSuperfluidMakerDAOGnosisLens ProtocolPOKT NetworkTellorLiquitySagaElizaOS
HeySynpressCerebrasGGMLRaid GuildMetagameScopeLiftAlluoDebtDAOMustang ProtocolPangeaNextMetaEthereum FoundationZKSyncGitCoinSuperfluidMakerDAOGnosisLens ProtocolPOKT NetworkTellorLiquitySagaElizaOSHeySynpressCerebrasGGMLRaid GuildMetagameScopeLiftAlluoDebtDAOMustang ProtocolPangeaNextMeta
HeySynpressCerebrasGGMLRaid GuildMetagameScopeLiftAlluoDebtDAOMustang ProtocolPangeaNextMetaEthereum FoundationZKSyncGitCoinSuperfluidMakerDAOGnosisLens ProtocolPOKT NetworkTellorLiquitySagaElizaOSHeySynpressCerebrasGGMLRaid GuildMetagameScopeLiftAlluoDebtDAOMustang ProtocolPangeaNextMetaTestimonials
"Sybil Solutions brings a powerful blend of talent, integrity, and genuine curiosity. When they take on a task you can relax, knowing that they'll see it through to completion to the highest standards. I recommend them without a heartbeat's hesitation."
A
Adrienne
CMO, Gnosis
Latest Insights
View all →Part 1: I ragged every conversation I ever had with AI
I ran a privacy preserving analysis on 809 conversations. The results were humbling, surprising, and exactly what I needed to see.
Part 2: Training My Own Coding Model — The SFT and DPO Pipeline
What happens when you fine-tune a 14B parameter model on your own coding conversations? Real numbers, real costs, and real frustrations.
Stay Updated
Subscribe for insights on AI systems, developer tools, and building in public.