Expert peer review of a Product Development Plan is one of the most important steps in drug development. The Product Development Plan shapes the strategy and direction of every drug throughout development. The expert review identifies gaps, challenges assumptions, and strengthens the decisions that guide the programme forward. A single review typically requires specialists across more than 11 functions, takes weeks to complete and costs about US$100k.

For innovators in low and middle income countries, where the disease burden is often greatest, access to that level of expertise can be out of reach. Oktopi wanted to make rigorous, senior expert review available at near zero marginal cost without reducing confidence in the result.

Mantel worked with Oktopi from strategy through to production, combining solution architecture, agentic AI engineering, and delivery. A multidisciplinary Mantel squad worked alongside Oktopi’s domain experts to turn their knowledge into a taxonomy of more than 1,400 critical development questions and toolkits and more than 100,000 guidance points.

OktoPilot uses Claude as its core reasoning engine, accessed through AWS Bedrock, within Oktopi’s broader LLM lifecycle management strategy. This strategy continuously evaluates foundation models against Oktopi’s taxonomy-driven review tasks. The Claude model family currently offers the best-performing models for OktoPilot’s workloads.

OktoPilot combines Claude’s reasoning capabilities with Oktopi’s expert taxonomy, agentic orchestration and evaluation framework to review documents and return structured, cited and verifiable feedback. Rather than generating responses from general knowledge, it assesses each document against the relevant expert-defined criteria and returns structured, cited, verifiable feedback.

That approach, built around structured review rather than free generation, was deliberate. Experts can trace each finding back to its supporting guidance, assess the reasoning, and validate the output. Prompt caching and parallelisation made it practical to reason over documents of up to 1,000 pages while keeping the marginal cost of each query low.

The work took the product from a cold start to a production ready MVP in four months. The platform is available globally across multiple AWS regions, supporting Oktopi’s growing customer base of biotech and pharmaceutical companies as the product scales. The same underlying review engine can now support additional document types, including Investigator’s Brochures and other expert development narrative documents, without rebuilding the approach from scratch.

OktoPilot changed the economics and pace of expert review. A process that previously took weeks and required a month of senior expert effort can now be completed in minutes, validated at senior expert level in 96.7% of assessments by a panel of global drug development experts in a blinded calibration study.

The work also supported Oktopi’s relationship with the Gates Foundation, whose funding reflects a shared commitment to expanding access for innovators in low and middle income countries.