TECHNOLOGY
What we build on.
We pick per job rather than per preference, and we work in your account with your tooling where you have it. This is the current working set.
01
Models and inference
Chosen by benchmark against your data and your cost ceiling, and swappable without rewriting the application.
- OpenAI
- Anthropic
- Meta Llama
- Mistral
- Google Gemini
- Self-hosted open weights
- Amazon Bedrock
- Azure OpenAI Service
02
Agent and retrieval frameworks
Orchestration is a means, not a religion. We use the lightest thing that survives production.
- LangGraph
- LlamaIndex
- Model Context Protocol
- Pinecone
- Qdrant
- Weaviate
- pgvector
03
Data and streaming
The unglamorous layer that decides whether anything above it can be trusted.
- Postgres
- Snowflake
- Databricks
- Apache Spark
- Apache Kafka
- Airflow
- dbt
04
Cloud and runtime
Built in your cloud account, under your controls, so nothing depends on our tenancy.
- AWS
- Microsoft Azure
- Google Cloud
- Kubernetes
- Docker
- Terraform
- On-premises and air-gapped
05
Vision and edge
Where the decision has to happen on the floor rather than in the cloud.
- PyTorch
- OpenCV
- YOLO
- NVIDIA Jetson
- Industrial cameras
- ONNX Runtime
06
Evaluation and operations
How we know a system still works six months after launch.
- MLflow
- Weights & Biases
- OpenTelemetry
- Grafana
- Custom evaluation harnesses
07
Application layer
The part people touch.
- TypeScript
- React
- Next.js
- Python
- FastAPI
- Node.js
- React Native
Cloud
Isotropic is an AWS Partner. We build on Microsoft Azure and Google Cloud as well, without a formal partnership on either. Work runs in your cloud account under your controls, not ours.
