AI Research
Three Tiers of Intelligence
We are a team of AI-native engineers and researchers pushing the boundaries of what's possible with language models.
Our Mission
We are a team of AI-native engineers and researchers.
Back when BERT was still a lab experiment, we were already pushing language models to trade the news milliseconds after the headline broke. Today we design and train hybrid-architecture models, curate multimodal data (text, video, timeseries... even LiDAR), and build the curricula that turn raw tokens into reliable intelligence.
What are we actually chasing?
We see three tiers of "hard" for an LLM. Let be the parameters of a model and , be the input/output:
Known Knowledge
The model already knows. Most standard chatbots live here—responding from memorized knowledge.
Contextual Reasoning
The model can figure it out—if we hand it the right context. Most agentic frameworks and RAG systems fit this description.
Adaptive Learning
The model itself must change to solve the problem. This is fine-tuning—adapting parameters to new domains or tasks.
This is reminiscent of complexity hierarchies, i.e. . And the frontier keeps moving. As models grow stronger, Tier-2 tricks are collapsing into Tier-1 intuition, and Tier-3 retraining is turning into Tier-2 prompting.
We’re here to accelerate that motion until the gaps disappear.
Close the context gap: make world fact inferable from the prompt alone.
Close the training gap: make new skills learnable in-context.
One token at a time.
Investment
Scientific Investment Approach
Signal-focused asset management powered by AI-driven timing and disciplined portfolio execution.
Intelligence Infrastructure
- •Unified data foundation with point-in-time integrity and lineage, integrating structured market/fundamental datasets and unstructured/alternative content
- •AI timing engine spanning macro, industry, and style with interpretable allocation outputs
- •Representation-learning pipelines and scalable real-time architecture from research to execution, with validation and auditability
Investment Methodology
- •Generate decision signals under regime context and calibrate horizons
- •Translate signals into policy- and risk-constrained allocation tilts and construct portfolios accordingly
- •Execute with cost and liquidity awareness, monitor outcomes, and adapt through structured feedback