Case study · AI startups
From LLM wrappers to agents
Design reviews and workshops for venture-backed engineering teams moving from a single LLM call to agents built on LangGraph, CrewAI or AutoGen.
- Client
- Portfolio companies of venture investors
- Result
- Several portfolio companies
- Stack
- LangGraph, CrewAI, AutoGen, LLM APIs, MLOps
The problem
Many early AI products are a prompt around an API. To hold up with real customers they need agents that plan, call tools, recover from errors and can be operated by a small team.
The work
- Reviews of agent orchestration, external API integration and error handling.
- Implementation guidance on LangGraph, CrewAI and AutoGen, with a focus on keeping systems modular and operable.
- Workshops and one-to-one sessions for the engineering team.
- Recommendations on cloud deployment and MLOps as the company grows.
Outcome
Each team leaves with an agent architecture it understands and can run on its own.
The client isn’t named, to respect confidentiality. I can share more detail on a call where permitted.
Talk about a similar projectMore work
Cross-asset research platform
Pipelines and Dash apps that give a sovereign wealth fund’s bonds and FX researchers one place to pull, compare and backtest market data.
In daily use by research and trading for 3+ years
Retention & cash analytics
Churn models, customer segments and branch cash forecasts for retail banks, from feature store to monitored production.
Led a team of four across several banks