Research
Inside χ-Bench’s healthcare workflows
A benchmark for complete healthcare workflows across providers and health plans.
Read the storyNews & Blog
Research, engineering, and ideas from Actava.ai.
Research
How Cura 1T learns through evaluation
How evaluation failures guide training data for a healthcare foundation model.
Read the storyProduct
Training specialized models with Model Factory
How verified workflow experience supports training and checkpoint selection.
Read the storyResearch
29 articles
Research2 min read
Following an AI improvement into the next round
A survey co-authored by Weiran Yao maps how changes to data, memory, tools, and learning procedures carry forward and how to measure their contribution to later improvements.
Research2 min read
DeltaFlow makes iterative language denoising more efficient
DeltaFlow combines bidirectional recurrent memory with noise-aware updates to improve language denoising, with experiments measuring generation quality and throughput at longer sequence lengths.
Research2 min read
How Cura 1T turns evaluation traces into better training data
Cura’s recursive self-improvement loop diagnoses failed trajectories, synthesizes targeted examples, and tests whether each training update preserves previously learned capabilities.
Research2 min read
Inside CHI-Bench’s simulated healthcare workflows
CHI-Bench places agents inside provider, payer, and care-management applications, then checks the persistent record of their decisions, artifacts, and handoffs.
Research2 min read
Giving language models a richer interface for customization
A position paper studies how vector prompts absorb supervision and influence model computation, then proposes exposing them as an additional interface for model customization.
Research2 min read
Preparing the next answer while the conversation continues
VoiceAgentRAG pairs background retrieval with a foreground responder, achieving an average cached retrieval time of 0.35 ms on cache hits in synthetic CRM conversations.
Research2 min read
AudioCapBench examines what multimodal models hear
Across 1,000 sound, music, and speech clips, AudioCapBench measures the correctness, coverage, and grounding of generated audio descriptions.
Research2 min read
Using diffusion models to refine a system prompt
A masked editing procedure refines instructions from interaction traces, improving tool-use and language-understanding results with a fixed GPT-4o-mini target.
Research1 min read
Following a large-codebase task with LoCoBench-Agent
An interactive software-engineering benchmark connects repository exploration, tool use, error recovery, and architectural consistency.
See what ACTAVA can do for you
Bring one workflow your team still runs by hand. We'll show you the agent that runs it.