When do compressed vectors make the same decisions?
Places quantization error inside the ranking and graph-pruning decisions that vector search actually makes, relating reliability to comparison margins, correlated residuals, and execution traces.
ο CETI / MIRATHE WONDERFUL STARARTISTIC INTERPRETATION
A WONDERFUL STAR / ASTRMIRA
Astrmira
Seek the unknown. Build beyond the horizon.
From first principles to real-world intelligence.
We study foundational questions in data, computation, and intelligence,
then turn rigorous ideas into systems people can use.
DATABASES✧RETRIEVAL✧VECTORS✧QUANTIZATION✧POST-TRAINING✧MATHEMATICS
OUR RESEARCH
From one question,
toward new understanding.
We frame questions, build methods, and test ideas through derivation and experiment—letting each finding become the beginning of the next inquiry.
RESEARCH, IN THE OPEN
Keep asking
into the unknown.
From foundational problems to new findings across different fields, this is where we record our papers and research progress.
Why does the value of the same quantization move change?
Studies the real utility of a discrete move in extreme low-bit LLMs by evaluating the move along its own path and combining decisions in an evolving model state.
What survives at one or two bits?
Explains low-bit ranking behavior through covariance structure and coordinate heterogeneity, including why an extra magnitude bit and random rotation help different representations in different ways.
QuIVer: building the search graph directly in binary space
Builds, prunes, and navigates an ANN graph in a training-free two-bit space, reading full vectors only for final reranking and exposing the data regimes where compact topology works.
PAPERS · FINDINGS · PROGRESS
OUR PRACTICE
Exploration goes
beyond the paper.
Foundational research expands what is possible. Building systems lets real problems answer back.
IDEAS, MADE TANGIBLE
Build research
into reality.
From data foundations and intelligent companions to sustained execution for complex work.
SYSTEMS
TriviumDB
An AI-native embedded database in Rust that unifies vectors, property graphs, and documents for agent memory and local knowledge.
INTELLIGENCE
infOS
An open-source runtime for AI companions, connecting long-term memory, context, and tools while powering persistent conversation and desktop interaction.
AGENT HARNESS
Piarium
A complete agent harness for complex engineering work, spanning code retrieval, context and memory, multi-agent coordination, and verifiable delivery.
A NOTE ON OUR NAME
A wonderful star.
A lasting curiosity.
Stars change. Curiosity continues.
Astr evokes the distant sky; Mira is a real variable star; Mirage is an imagined horizon still to be reached. We call this image 幻梦星芒—a wonderful star.
Mira changes in brightness and leaves a long ultraviolet tail. It reminds us to preserve imagination around the unknown and leave a trail that others can continue.
Name inspiration · NASA / The Tail of a Wonderful Star ↗LET’S ASK THE NEXT QUESTION
The next question worth studying
may come from you.
Databases and intelligent systems · Research collaboration · University partnerships
Begin a meaningful collaboration with one real problem.