Customer Highlight
How teams build, ship and scale their AI products on Brightmoor infrastructure.
Halden’s Kestrel and Brightmoor
Industry:Consumer internet
Use case:High-volume image and video generation via inference
Results:46% of Kestrel’s media messages, faster responses, higher satisfaction
46%
of Kestrel’s image and video messages powered by Brightmoor
31%
faster response time compared to their previous provider
22%
increase in positive feedback on Brightmoor-powered multimedia bots
37%
of Kestrel’s official generation bots now run on Brightmoor
Challenge: Kestrel needed faster, higher-capability media results at growing volume
Generative models for image and video are large and compute-heavy, which pushes latency up during inference. Kestrel needed a partner that could keep responses quick and high-quality without trading away model accuracy.
- Optimise inference for large generative models.
- Keep access to the newest image and video models current.
- Scale reliably against a fast-growing request volume.
Solution: Partnering with Brightmoor
- Reduce inference latency: an optimised pipeline cut response times sharply for Kestrel’s media bots.
- Access newer models: Kestrel now selects from a continuously refreshed model library.
- Handle peak volume: robust API endpoints absorb traffic spikes without degrading the experience.
“We’ve been struck by the speed and scalability of Brightmoor’s inference. Quick access to newly released models keeps our product moving with a market that changes weekly.”
Marisol Trejo
Product Lead, Kestrel by Halden
Outcome: 46% of Kestrel’s media messages, faster responses, happier users
- Message share: Brightmoor bots account for ~46% of generated media on Kestrel in Q1.
- Like ratio: those bots receive 22% more positive feedback than other multimedia bots.
- Speed: response times are 31% faster than the previous provider.