Hakase AI launches an operating system for drug development

Hakase AI introduced an AI-native operating system for drug development that uses one engine, one evidence record and one governance model across seven applications. The company also named a new CEO and CTO as it pushes an operating model designed to preserve provenance, review history and accountability across pharmaceutical R&D.

Why it matters: - Drug-development teams often lose evidence lineage as AI outputs move from discovery to pre-clinical, clinical, regulatory and post-market work. - Hakase AI is trying to fix that with a shared operating layer that keeps source versions, transformations, uncertainty signals and reviewer history attached to each output. - The approach is meant to make AI outputs more reviewable, reproducible and usable in regulated workflows.

What happened: - Hakase AI introduced an AI-native operating system for the drug-development lifecycle on Sept. 17, 2026. - The platform uses a common engine, a shared evidence record and one governance model instead of a separate AI stack behind each tool. - Seven applications run on the engine today. - Four of those applications were built by AKT Health, Hakase AI's product partner. - Hakase AI also named Dr. Manasa Kondamadugu as chief executive officer and Pranay Dinavahi as chief technology officer.

The details: - The Hakase Engine is an evidence-backed multimodal orchestration layer. - The engine routes work across model families suited to molecular representation, protein sequence, structure prediction, clinical text and physiology-based solvers. - The engine grounds outputs in versioned retrieval from named public sources. - The engine records inputs, source identity, source version, transformations, model and configuration, uncertainty signals where supported, and the reviewer who accepted the result. - Applications running on the engine inherit those services instead of rebuilding them. - Each application remains responsible for its intended use, accepted inputs, outputs, controls and stated limitations. - The seven applications cover pre-clinical computational assessment, clinical design analysis, source-grounded scientific drafting, regulatory and safety operations, clinical operations data capture, real-world evidence and business-development intelligence. - Three of the applications were built by Hakase AI. - Four were built on the same engine by AKT Health and remain AKT Health products. - A separate discovery partner supplies candidate structures into the earliest computational workflow. - Hakase AI says the third-party applications are a practical test of the architecture because the runtime, evidence model and controls must hold for software the company did not write. - Pranay Dinavahi said the key problem is determinism of the record, not model quality. - Dinavahi said outputs must carry retrieval sets, source versions, model and configuration identifiers, transformations and reviewer information so results can be reconstructed later. - Dinavahi said the engine propagates uncertainty rather than collapsing it into a score and keeps evaluation and reviewer gates at the boundary. - Dr. Manasa Kondamadugu said the operating system provides the environment, controls and continuity for accountable people to make decisions. - Kondamadugu said early-development teams often lack the context needed to make a result reviewable months later by someone not present for the original analysis. - Kondamadugu said Hakase is built so outputs arrive with sources, assumptions and limitations attached. - The engine is model-agnostic and can reroute tasks as new models become available. - Retrieval is grounded in public sources covering target biology and pharmacology, chemistry and bioactivity, protein structure, human expression and genetic variation, toxicology, clinical trial precedent, and regulatory and post-market safety. - Source coverage, licensing and access conditions vary. - Source consultation does not make an output complete, current or validated. - Hakase follows a predict-then-confirm approach. - Computational outputs are for hypothesis generation and prioritization. - Laboratory work is still required to confirm observations. - Agreement between computational methods is not confirmation. - Outputs are decision-support materials and are not dose recommendations. - Outputs do not determine a protocol or establish safety, efficacy, trial results or regulatory acceptability. - Hakase uses the term HAIOps for its own operational governance approach. - HAIOps is not an external standard, certification or attestation. - Provenance records, evaluation, uncertainty presentation and reviewer gates are implemented by workflow and configuration. - Those controls support assessment and do not transfer responsibility from the user. - Workflows are designed against relevant ICH guidance for nonclinical safety, safety pharmacology, genotoxicity, model-informed drug development and good clinical practice. - Hakase also aligns the workflows with established frameworks for model credibility. - Hakase says alignment with a framework is a design consideration, not certification or regulatory acceptance. - Hakase publishes its full source register, framework list and lifecycle coverage map at hakase.ai. - The applications are available for scoped technical evaluation. - Scope, inputs, outputs and validation status are agreed per engagement. - Enquiries go to info@hakase.ai.

Between the lines: - Hakase is pitching an operating-system model, not a point-solution model, to solve a core pharma problem: evidence breaks when tools do not share provenance and controls. - The partner-built applications matter because they show the platform is designed to support third-party software under the same governance layer. - The company is also drawing clear limits around what the platform does and does not do, likely to fit regulated workflows without overclaiming clinical or regulatory authority.

What's next: - Hakase AI will continue offering the described applications for scoped technical evaluation. - Each engagement will define scope, inputs, outputs and validation status. - The company is positioning HAIOps, its source register and its lifecycle coverage map as part of the evaluation framework for prospective users.

The bottom line: - Hakase AI is betting that drug development needs one governed evidence layer more than another standalone AI tool.