Research

About SeraYun Research

SeraYun is an independent development effort exploring how AI systems can retain useful context, explain their decisions and support real work. The portfolio combines local prototypes, controlled experiments and field-oriented software development. Business registration is in preparation. This page does not claim incorporation, commercial readiness, sponsorship or selection for a support program.

Status reviewed 14 October 2026. Implementation checks, public-page observations and plans are distinguished below. No private conversations, manuscripts, trading records or internal datasets are published here.

Research Areas

• Evidence-aware memory

  SeraLab explores local long-term memory, source attribution, retrieval, revision and withdrawal. The aim is inspectable recall rather than treating retrieved text as unquestionable truth.

• Local and multi-model systems

  Sereaniel provides an application context for local LLMs, multiple models and agentic workflows. Research questions include context handoff, permissions and the boundary between conversation and durable work.

• Reproducible strategy evaluation

  Sera Trader, CoinLab and StockLab explore automated research, strategy evaluation, isolated PAPER or SHADOW simulation and decision tracing. Simulation is not live execution, profitability evidence or investment advice.

• Long-form writing workflows

  Sera Author explores world consistency, multi-model proposals, master and episode plans, manuscript versions and resumable work. Generation remains conditional on model, budget and exact master approval; a completed novel has not been demonstrated.

• Digital safety and field workflows

  SafeSM is a field-oriented development case for digital safety records and business workflow automation. Public interface availability is distinct from authenticated workflow acceptance and production use.

Projects & Experiments

SeraLab — primary local research case

Local storage, retrieval and source-linked version handling are implemented. Synthetic import, duplicate handling, revision and recovery checks have been performed, along with a local hybrid-retrieval run. A full evaluation on a supplied personal export and a public retrieval-quality benchmark remain outstanding. Project overview.

SafeSM — primary field-development case

A public site has been observed in earlier checks. This is not evidence that login, record entry, reports or field workflows are fully validated. A reproducible, non-sensitive demonstration and acceptance criteria are the next evidence targets. Project overview.

Sereaniel & Sera Author — application studies

Local conversation and independent workspace entry are connected. Author supports separate worlds and works, side stories, editable plot/manuscript references and version history. Isolated browser tests cover edits, work separation, pause/resume and restart persistence. Native desktop automation and actual provider output quality are separate, limited validation areas. Sereaniel overview.

Trader, CoinLab & StockLab — research and simulation only

Recorded local tests exercise authenticated research, durable requests, saved receipts and isolated simulation. NO_TRADE is a possible valid outcome, not proof of a fill. Existing credentials, market-data freshness checks and risk gates constrain execution. No live-trading return is reported. CoinLab · StockLab.

Media and statistical experiments

LottoLab studies statistical evaluation and reproducibility, without promising prediction or winnings. Studio is an experimental local media workflow. Public descriptions do not imply commercial releases. LottoLab overview.

Technical Reports

These public engineering notes describe methods and limitations. They are not peer-reviewed papers, performance benchmarks or exports of private evidence.

• TR-001: Source-aware memory and revision checks
• TR-002: Durable requests and simulation boundaries

Development Updates

14 October 2026 — Research documentation and a source-controlled website workflow introduced. Local workspace entry, plot-reference versioning and distinct service icons have been checked. The public site now connects the Labs, Research and Original Worlds in one continuous homepage, with source checks and automatic deployment from the repository.

Contact & Collaboration

Interested in reproducible evaluation, local AI systems, memory reliability or practical safety workflows? Contact founder@serayun.com. This is the existing public contact address. Please do not send credentials, personal records or confidential datasets in an initial message.