B
Bespoke Labs Tech Stack
AI infrastructure for reinforcement learning and LLM post-training
Bespoke Labs builds infrastructure for RL-based AI agents and LLM fine-tuning, with a tech stack optimized for GPU-accelerated training (CUDA, PyTorch, vLLM, Triton). The project list reveals dual focus: core platform work (agentic AI, RL environment curation, CUDA kernel development) alongside applied work (trading systems, backtesting, adversarial analysis). Pain points cluster around scaling research to production and managing frontier-scale agent training — a gap their 11-engineer, 4-researcher mix is actively hiring to close.
Tech Stack 80 technologies
Core StackPyTorch Python NumPy pandas scikit-learn Apache Airflow Prefect AWS TypeScript Next.js PostgreSQL Elasticsearch Splunk CrowdStrike CUDA vLLM Nsight Systems Nsight Compute Triton NVLink NCCL C/C++ SciPy JAX statsmodels GCP Azure Bloomberg MITRE ATT&CK Bash+49 more
What Bespoke Labs Is Building
◆Challenges
- Building production-grade rl infrastructure
- Adversarial risk in ai models
- Supporting agent training at frontier scale
- Scaling data pipelines
- Multi-million dollar contracts
- Scaling research prototypes to production
- Large-scale deployment of rl environments
- Translating research insights to production
- Model performance improvement
- Complex workflows for rl environment curation
▲Active Projects
- Calibration, inversion, deconvolution, and uncertainty quantification pipelines
- Develop systematic trading strategies
- Build trading systems
- Implement backtesting frameworks
- Core platforms behind agentic ai
- Cuda kernel development for ai training and inference
- Adversarial analysis and scenario planning
- Gpu profiling and benchmarking tooling
- Customer project delivery
- Data pipeline scaling
Hiring Activity
20 roles · 20 in 30d
Department
Seniority
Notable leadership hires: Tech Lead
Company intelligence