Stacks/Enterprise research

Grid Dynamics deep-research agent (Fortune 500 manufacturer)

Grid Dynamics built a deep-research agent for a Fortune 500 manufacturer and published a candid post-mortem: LangGraph got them to production, but custom retry and Redis state management became expensive to support. They migrated to Temporal for durable execution, retries, and human-in-the-loop waits, and had to rethink how API clients and configuration are managed inside activities.

Use case

Deep research across a Fortune 500 manufacturer's internal databases, shared drives, and local repositories, expanding to the open web with citations when internal data is not enough, for teams at 100+ plants

Category

Enterprise research

Team size

not stated

Orchestrator / framework

The team first took a LangGraph prototype to production, then migrated to Temporal after production pain around error handling, retries, human-in-the-loop waits, and custom state. The earlier system used Apache Kafka so user requests landed on a queue and were consumed by a pool of executors, aiming for exactly-once processing.

Memory

LangGraph state was kept in Redis, which the authors say created lifecycle, expiration, and cache-debugging problems. After the move, Temporal owns durable workflow state. Activities are self-contained; the team used client pooling and lazy initialization instead of re-initializing API clients in every activity.

Evals

not stated

Observability

not stated

Monthly cost

not stated

Models

not stated

Observability tools

not stated

Other tools

Redis (prior state store)Apache Kafka (prior request intake)

Source

https://temporal.io/blog/prototype-to-prod-ready-agentic-ai-grid-dynamics

Dmitry Mezhensky, Dmitry Larko, and Eugene Steinberg (Grid Dynamics) · Sep 28, 2025