TWOG

Agent-assisted research infrastructure

TWOG - A Living Research Engine

TWOG Autoresearch is building a source-traceable research platform for canine hemangiosarcoma and related human translational oncology. It ingests biomedical evidence, organizes it into a searchable knowledge library, and gives specialist agents the tools to search, compare, design, test, and evaluate research paths for human review.

Built for evidence organization, hypothesis support, validation planning, and human-reviewed research workflows. Not for diagnosis, prescription, or treatment recommendations.

PapersCompoundsProteinsMolecular dataLeads
Evidence Librarychunksentitiesclaimscitations
ScoutHypothesisSkepticToolsReview

Why it exists

The evidence is fragmented.

Canine hemangiosarcoma is aggressive, under-served, and biologically complex. Relevant evidence is scattered across veterinary oncology, human angiosarcoma, vascular tumor biology, sarcoma literature, drug databases, protein references, molecular datasets, safety records, and emerging online signals.

PubMedEurope PMCPMC OAOpenAlexChEMBLPubChemUniProtRCSB PDBOpenFDAMolecular dataX/Twitter leadsPartner data

The platform

From evidence to research workflows.

TWOG Autoresearch continuously ingests research sources, structures them into an evidence library, prepares them for retrieval, and gives specialist agents access to the information and tools needed to generate source-traceable research artifacts.

01Ingest
02Structure
03Retrieve
04Agent Work
05Scientific Tools
06Human Review

Evidence library

Raw records become usable research context.

Records are normalized into research objects, document chunks, entities, claims, citations, and embeddings. That library supports RAG, MCP tools, source review, citation-backed briefs, and future scientific workflows.

evidence_libraryretrieval_ready
Research objectspapers, records, datasets
Document chunksretrieval-ready context
Entitiesgenes, proteins, compounds
Claimssource-linked assertions
Citationstraceable evidence paths
EmbeddingsRAG-ready search layer
claim_0421VEGFR signaling appears in canine HSA and related vascular tumor literature.[C12] [C18]

Research agents

Agents do bounded research work.

Specialist agents operate on top of the evidence library. They search across sources, compare findings, identify candidate mechanisms, design hypotheses, plan validation work, inspect contradictory evidence, and produce source-traceable reports for human review.

Evidence Scout

Finds relevant records, gaps, contradictions, and source follow-up targets.

ready
Hypothesis Designer

Connects mechanisms, pathways, compounds, and translational signals.

ready
Skeptic Review

Challenges weak evidence, unsupported claims, and overconfident conclusions.

challenge
Validation Planner

Turns promising directions into reviewable experimental or computational plans.

ready
Scientific Tool Runner

Routes selected tasks into modeling, molecular data, safety, or GPU-backed workflows.

queued

Scientific tools

The frontier layer connects agents to computational research.

The goal is not to lock the project to one tool stack. The durable layer is a controlled interface where agents can hand off evidence-backed questions into reproducible computational workspaces, capture inputs and outputs, preserve provenance, and route results back through human review.

Controlled workspaces

Reproducible inputs

GPU-ready execution

Structured outputs

Provenance capture

Review gates

Validation artifacts

Hypothesis pipeline

Ideas move through gates, not shortcuts.

The system is designed to move from evidence to hypotheses in a controlled sequence: ingest sources, structure evidence, retrieve relevant context, generate multi-perspective briefs, evaluate synthesis quality, identify candidate hypotheses, plan validation steps, run appropriate tools, and route outputs to human experts.

01Evidence
02Brief
03Evaluation
04Hypothesis
05Validation Plan
06Tool Run
07Expert Review

Quality and oversight

Every output should remain traceable.

Research briefs require citations. Synthesis outputs are evaluated for citation coverage, perspective balance, contradiction handling, novelty, actionability, and readiness for review. Human oversight remains central, especially for clinical interpretation, research prioritization, and validation decisions.

Citation coverage

Perspective balance

Contradiction handling

Novelty

Actionability

Human review readiness

Current build status

Active research infrastructure, built in public.

V1 proved the major lanes could work. V2 is rebuilding them with cleaner orchestration, storage, retrieval, provenance, and review contracts. This is not a medical device, diagnostic system, or treatment recommendation engine.

V2 active build

Ingestion bridgeSource monitoringFull text hardeningEmbeddingsResearch briefs

V1 experimental baseline

Lead follow-upAgent review workflowsSource-depth explorationSynthesis evaluationGPU validation pathwayModeling and molecular-data prototypesPartner validation planning
Graffiti the corgi

Why this work exists

For Graffiti, Brady, and every dog after them.

TWOG began because hemangiosarcoma leaves too many families with too few options and too little organized evidence. The platform is being built to make the research easier to search, compare, challenge, and move responsibly toward validation.

Collaborate

Help improve the evidence, review the work, or shape the validation path.

TWOG Autoresearch needs researchers, veterinarians, technical contributors, data contributors, partners, and supporters who can help turn scattered evidence into reviewable research artifacts.