Kreflux
Reasoning at scale
Intelligence should not be a black box with arbitrary compute constraints. Users deserve transparent, adjustable effort levels and developer tools that seamlessly bridge the web, the terminal, and cloud APIs.
Personal bio and project archive
Jayden Xing
Founder of Kreflux, debater, and builder behind XingTalk. I use XingTalk to collect the work I am most serious about: independent AI reasoning models with Kreflux and Freezesol, competitive debate, cultural writing, and learning in public.
- Founder of Kreflux
- Freezesol Alignment
- Competitive speaker
- Debater and writer
The record / Lincoln-Douglas, 2025 to 2026
Ballots don't lie.
Every round tracked across the season, from Equality in Forensics scrimmages to the CHSSA Middle School State Championship. The constellation is the whole season: one star per tournament, one moon per round, read clockwise from September.
- Overall record
- 71.9%
- Matches won
- 82
- Prelim rounds
- 70.3%
- Elim ballots
- 62.5%
Each tournament becomes one note, pitched by its win rate. Tap any star in the constellation to hear its rounds: wins climb the scale, losses fall.
Tabroom.com / Team record 1705063
The official record lives on Tabroom.
Every result, ballot, and speaker point from the season, entered by tournament staff and updated as new rounds post. The figures above come straight from it. Tabroom asks for a free login before showing results.
Tournaments
25
Elim runs
8
Best streak
12
Wins Losses
Season ledger
Showing all 25 tournaments, newest first.
- Middle2.5-1.55-1State finalist
- Open0-10-00-1 Record
- Middle6-03-1Quarterfinalist
- Novice3-20-03-2 Record
- Middle4-10-1Octofinalist
- Open1-00-01-0 Record
- JV3-20-03-2 Record
- Open3-30-03-3 Record
- Open1-00-01-0 Record
- Middle1.5-1.50-01.5-1.5 Record
- Open3-30-03-3 Record
- Open1-00-01-0 Record
- Open1-30-01-3 Record
- Open2-00-02-0 Record
- Middle4-00-04-0 Record
- JV6-01-1Octofinalist
- Open0-10-00-1 Record
- Novice3-24-1Runner-up
- Open1-00-01-0 Record
- Novice4-21-1Quarterfinalist
- Novice4-22-1Semifinalist
- Middle3-10-03-1 Record
- Open1-00-01-0 Record
- Open5-02-0Champion
- Open1-00-01-0 Record
Win Loss Split
Resource
LD Debate Crash Course
An introductory guide to Lincoln-Douglas debate basics, perfect for beginners. Covers fundamental concepts, basic argument structure, and debate terminology. PowerPoint presentation, 11 slides.
Tabroom
Ballots, judges, and speaker points.
The ledger above mirrors the season. For judge decisions, speaker points, and anything posted after this page was updated, the Tabroom record is the source of truth.
Featured systems
Reasoning models & alignment.
Artificial Intelligence and Safety Division. Independent AI research by Jayden Xing, building adaptive language models at Kreflux and verifiable defenses against jailbreaks, extraction, and abliteration at Freezesol.
kreflux.com / Live platform and lab
KrefluxStay curious. Reasoning at your pace.
An independent AI lab building reasoning language models and tools with user-controlled thinking effort. From the browser chat to developer terminal workflows and API pipelines, Kreflux gives users real-time control over how deep the model reflects on complex problems.
Adjustable reasoning effort
Balanced for everyday reasoning.
Reasoning tree at3levels of thought
klucidKlucid CLI
A high-performance command-line autonomous agent for developers, directly integrated with Kreflux models and terminal workflows.
kreflux apiKreflux API
Programmatic access to Kreflux models for products and pipelines, with the same adjustable reasoning effort as the chat.
git clone github.com/Kreflux/openkrefluxOpenKreflux
The open side of the lab: models and tooling published on GitHub for anyone to inspect and build on.
freezesol.com / Alignment division
Freezesol
A sun, held in stillness. Permissions are enforcement.
The alignment and safety division of Kreflux. Because text defenses are merely advisory and prompts can be bypassed, Freezesol builds verifiable permission barriers, extraction defenses, and safeguards engineered to survive intentional model abliteration.
Interact
The boundary tightens, the plasma stops, and the soundtrack drops into a single held chord.
Krextract
DefenseScreens extraction attempts and gives applications an explicit, inspectable boundary between private model reasoning traces and public answers.
npm install github:Kreflux/KrextractKreGuard
LayerOpen guardrail layer for LLM applications. Screens inputs, outputs, and egress for jailbreaks and prompt injection with zero runtime dependencies.
pip install git+https://github.com/Kreflux/KreGuardRSAAJ
ResearchRobust Safeguards Against Abliteration and Jailbreaks. Training models to preserve core refusal behaviors even when refusal activation directions are subtracted.
The Frozen Sun
Short filmA sixty-second visual exploration on the balance between sheer computational power and deliberate safety boundaries.
Philosophy
Capability & alignment.
State-of-the-art AI requires two complementary forces working in tandem: deep exploratory reasoning power and rigorous, unyielding safety guardrails. Drag to balance them, and watch the binary and the fabric of spacetime beneath it respond.
Alignment50%
Capability50%
Stable binary
Both stars hold each other in orbit. This is the point of the whole lab.
Freezesol
Enforcement, not polite advice
Prompt instructions and semantic rules alone cannot prevent sophisticated jailbreaks or weight abliteration. Safeguards must be structural, permission-based, and mathematically resilient under adversarial conditions.
Signal
The through-line is clarity.
Good debate, good writing, and good tools all depend on turning complexity into something people can actually use.
Strategy
Argument structure, evidence, speaking decisions, and competitive reflection.
Systems
AI projects, interface experiments, and practical technical learning.
Story
Personal writing about culture, identity, and the meaning behind the work.
Bio snapshot
A profile built around communication and technical curiosity.
I am an aspiring student, developer, and competitive speaker focused on exploration, research, and technical growth. My core interests lie at the intersection of technology and communication: the logical systems of Artificial Intelligence and the rigorous challenge of Speech and Debate.
The site is a living record of that process, from tournament results and AI lab projects (Kreflux and Freezesol) to experimental interface modules.
Current focus
Turning messy ideas into clear arguments and usable projects.
The name
Xíngrising tone
Xing is 邢, not 星 (xīng, star). The 阝 on its right is the 邑 radical, the old sign for a walled city, and the surname traces back to the State of Xing under the Zhou dynasty, in what is now Xingtai, Hebei.
So this galaxy is not a pun on a star. It is a map of settlements: every tournament, lab, and essay is a city of its own, and you are the one traveling between them.
Explore
Choose a part of the biography.
Each section opens a different side of the work, from competitions to experiments. Every planet in this system is one of them: hover a card or a world to link the two.
Back the mission
Directly support debate research.
Help support the projects, tournament travel, and ongoing learning through one secure Stripe checkout. The pulsar keeps time while you decide.
Secure checkout on Stripe
On pulsar time. Sound on to hear it.
Where it goes
Tournament travel
Entry fees, travel, and lodging for a full Lincoln-Douglas season.
The projects
Kreflux, Freezesol, and the experiments that start out in the Lab.
Learning in public
Keeping XingTalk, its resources, and its experiments free to read and use.