How we work
Small steps, real results
We start small, show something working early and grow it only when it has proven its value. This is how an engagement with us runs.
The process
Four steps, each with a clear result
At the end of every step you hold something concrete and decide whether to continue. The times are typical, not fixed.
- 011–2 weeks
Discover
We learn your process, your data and your goal. We tell you honestly whether AI is the right tool.
You receive
A short written assessment and a proposal.
- 022–4 weeks
Prototype
We build a small working version on your real data.
You receive
A working demo and measured results.
- 034–12 weeks
Build
We turn the prototype into a production system inside your environment.
You receive
A deployed system with tests, evaluations and documentation.
- 04Ongoing
Run and improve
We monitor, measure and extend the system, or hand it over to your team.
You receive
Support, regular reports and training.
Ways to engage
Three ways to work with us
Fixed project
A defined scope, a fixed price and a delivery date.
Best when the goal is clear.
Embedded team
Forward AI engineers inside your team, on a monthly basis.
Best when the work changes as you learn.
Advisory
A review, a strategy or a readiness assessment.
Best when you need a clear view before you invest.
Principles
What you can expect from us
- Start small, then scale
- One task or one prototype first. We grow it only after it has shown its value.
- A person stays in control
- Important decisions wait for a human. You choose where that line is.
- Your data stays yours
- We work in your environment when you need it and use only the access you give us.
- We measure before we launch
- Every agent and every system has evaluations before it goes live.
- You own what we build
- Code, skills, evaluations and documentation are handed over to you.
- We say no when AI is not the answer
- Sometimes a simple script or a better process is enough. We will tell you.
Tools
Tools we work with
We build in Python. Agents are typed, tool-using programs with an evaluation set, traces and a person in control of the decisions that matter. We use the model providers you already trust, and we run open models when the work has to stay in your environment. Retrieval sits on a database you own. The system is deployed in your cloud.
- Language
- Python
- Agents
- LangGraph, Pydantic AI, the OpenAI Agents SDK, the Claude Agent SDK, Model Context Protocol
- Models
- OpenAI, Anthropic, Google, and open models (Llama, Mistral, Qwen) through vLLM, Ollama or Hugging Face
- Retrieval
- PostgreSQL with pgvector, plus a document parser
- Evaluation
- An evaluation set in the repository, RAGAS or DeepEval, traces in Langfuse
- API
- FastAPI
- Data
- dbt, Airflow or Dagster, Spark and Kafka, on Databricks, Snowflake or BigQuery
- Infrastructure
- Terraform, Ansible, Docker, Kubernetes and GitHub Actions
- Cloud
- AWS, Azure and Google Cloud, in your account
Models and tools are chosen per engagement. The cloud we use is the one you already run.
Questions
What clients ask first
How do we start?
With a conversation. You tell us about the process or the product, and we tell you if and how we can help. If it fits, the Discover step follows.
Do you work with our existing tools and cloud?
Yes. We build in the environment you already use, and we choose the models and platforms that fit your case.
What happens to our data?
It stays yours. We work inside your environment where you need it and use only the access you give us.
Who owns the result?
You do. Code, skills, evaluations and documentation are handed over to you.
Does a small first project make sense?
Yes, and we prefer it. One task or one prototype is the fastest way to see real results.
Ready for the first step?
It starts with a conversation. Tell us about the process or the product you have in mind.
Talk to us