MedSwin grounds clinical answers in patient records and guideline evidence.
+
MedSwin combines patient-specific EMR retrieval, clinical practice guidelines, biomedical reranking, specialist critique, and provenance tracking so each answer can be traced back to the evidence that shaped it.
-
-
- Research prototype — not a substitute for professional medical advice.
-
-
-
+
+
+
+
Live evidence sonar
+
Clinical channels under audit
+ 4,895 m
+
+
+
+
+
+
+
MedSwinevidence lock
+
+
+
+
+
+
Hover a beacon to inspect an evidence channel.
+
+
0B student
+
0B teacher
+
0agents
+
0% Recall@10
-
-
-
-
-
-
-
Overview
-
- MedSwin frames clinical QA as an evidence-constrained decision pipeline.
- Every answer is gated by evidence sufficiency, bounded by a strict context budget, and accompanied by a
- replayable trace suitable for audit and safety review.
-
-
-
-
-
- Answer
-
-
- Clinically phrased, uncertainty-aware output generated only when evidence gates are satisfied.
-
-
-
-
- Evidence bundle
-
-
- Compact EMR + guideline passages selected under token and diversity constraints.
-
MedSwin separates retrieval, evidence quality, safety critique, and answer synthesis.
+
The system tracks provenance, calibrated relevance, facet sufficiency, contradiction handling, agent reliability, benchmark performance, and release artifacts as distinct parts of the clinical evidence workflow.
+
+
+ 01
Evidence-first generation
Answer generation only follows an accepted EMR/CPG evidence bundle, rather than raw top-K context.
+ 02
Calibrated inclusion
Reranker logits are converted to calibrated probabilities for threshold-based policy checks.
+ 03
Safety is a first-class facet
Contraindications, interactions, exclusions, and contradictions remain visible throughout selection.
+ 04
Auditability over confidence
Each claim keeps source metadata, score context, evidence grade, and facet role.
-
-
-
-
-
-
-
-
1) Auditable Multi-Agent Orchestration
-
-
-
- Every agent produces artifacts with provenance metadata and logs tool calls + selected evidence. This creates a
- structured audit trail suitable for review and replay.
-
- Stage-1 retrieves candidates via dense retrieval + BM25 union. Stage-2 reranks with a long-context biomedical reranker
- (LoRA-adapted) and outputs calibrated probabilities used by policy constraints.
-
-
-
Hybrid candidate pool: ANN + lexical coverage
-
Evidence sufficiency thresholds for EMR/CPG
-
Budgeted, diverse selection (MMR-style)
-
-
+
+
+
Architecture animation
+
Trace a clinical query through the evidence chamber.
+
A clinician query is decomposed into patient context and guideline evidence, reranked with biomedical relevance signals, checked for sufficiency, and returned with a grounded answer plus audit trail.
+
+
+
+
+
Pipeline playback
+
Clinician query enters the chamber.
-
-
-
-
-
3) Distilled 7B Medical LLM Pipeline
-
-
-
- A compact student is built with SFT on augmented biomedical QA, then refined using hard+soft KD from a larger instructor.
- This targets deployability while preserving calibrated reasoning behavior.
-
-
-
Large-scale augmentation with semantic checks
-
Hard labels expand coverage; soft labels preserve uncertainty
-
PEFT (QLoRA/LoRA) enables modest GPU training
-
-
+
+
+
-
-
-
-
-
From question → audited answer
-
- MedSwin outputs an answer only when the evidence bundle is sufficient. Otherwise it asks clarifying questions or expands retrieval.
-
-
- Clarify
- Retrieve-more
- Safe final
-
-
-
-
-
-
-
1 Normalise
-
Canonicalise terms, expand abbreviations, form retrieval probes.
-
-
-
2 Retrieve + Rank
-
Hybrid candidates, rerank with calibrated probabilities, enforce sufficiency.
-
-
-
3 Synthesize + Critique
-
Summarise EMR, synthesise guideline actions, run safety critique, then answer.
-
-
-
+
+
+
-
-
-
-
-
-
Architecture Explorer
-
- Explore MedSwin layers: multi-agent workflow, two-stage retrieval, evidence sufficiency checks, and MAC coordination.
-
-
-
- Orchestrated
- Provenance
- Budgeted
-
-
-
-
-
-
-
-
+
+
+
Medical specialist
+
Training, distillation, and model merging are displayed as separate mechanisms.
+
The specialist model is built from supervised biomedical instruction data, teacher-student distillation, and training-free merge operators that control destructive interference between updates.
-flowchart TB
- Q["Query q + patient context"] --> O["MAC Orchestrator"]
-
- subgraph A["Specialised Probes"]
- E["Exploration Agent\n(recall-normalised)"]
- P["Policy Agent\n(guideline coverage)"]
- S["Safety Agent\n(contraindications)"]
- end
-
- O --> E
- O --> P
- O --> S
-
- E --> CE["Candidate C_e + scores"]
- P --> CP["Candidate C_p + scores"]
- S --> CS["Candidate C_s + scores"]
-
- CE --> U["Weighted union + re-rank\nω_i · S_i(q,d)"]
- CP --> U
- CS --> U
-
- U --> M["Meta-bundle M_MAC\n(token budget B + sufficiency checks)"]
- M --> OUT["Evidence for synthesis + critique"]
-
-
-
-
-
-
Why MAC helps
-
- Instead of a single retrieval hypothesis, multiple agents explore different constraints and then fuse evidence,
- preserving disagreements as inspectable signals.
-
-
-
-
Agent weights
-
- Agents accumulate track-record weights (e.g., sufficiency compliance) to influence future selection in a transparent way.
-
-
-
-
-
-
-
-
Single-Turn Sequence
- Retrieval → policy checks → synthesis
-
-
-sequenceDiagram
- autonumber
- participant C as Clinician
- participant O as Orchestrator (MAC)
- participant QN as Query Normaliser
- participant IR as Two-stage Retrieval
- participant EMR as EMR Summariser
- participant GL as Guideline Synthesiser
- participant SC as Safety Critic
- participant L as LLM (Distilled 7B)
-
- C->>O: query + patient context
- O->>QN: normalise + expand probes
- QN-->>O: canonical query
- O->>IR: retrieve candidates + rerank
- IR-->>O: evidence bundle M + p_cal + provenance
- O->>EMR: build clinical state summary
- O->>GL: extract actions + contraindications
- O->>SC: check evidence sufficiency + safety
-
- alt Insufficient evidence
- SC-->>O: retrieve-more / clarify request
- O-->>C: request clarification / expand search
- else Sufficient evidence
- SC-->>O: ok + cautions
- O->>L: synthesize answer with citations
- L-->>C: answer + citations + trace
- end
-
-
-
-
+
+
Hybrid SFT + KD
Teacher-student transfer
QLoRA
+
+
27BTeacher
+
hard labelstop-k soft logitsuncertainty transfer
+
7BStudent
-
-
-
+
Lt = α CE + (1 − α)τ² KL(pT(·|τ) ∥ pS(·))
-
-
-
-
-
-
-
-
- Two-Stage Retrieval & Calibrated Reranking
-
-
- Evidence selection is separated into recall-oriented candidate generation and precision-oriented reranking.
- This avoids early truncation while enabling deterministic, policy-aware inclusion decisions.
-
-
-
-
-
- Stage 1 — Candidate generation
-
-
- Dense ANN retrieval is unioned with BM25 to preserve rare clinical terms, abbreviations, and lab-specific phrasing.
-
-
-
-
-
- Stage 2 — Long-context reranking
-
-
- A biomedical LLM reranker scores each passage and outputs calibrated probabilities usable as policy thresholds.
-
-
-
-
-
- Policy-aware selection
-
-
- Final selection enforces EMR + guideline sufficiency, diversity (MMR-style), and a strict token budget.
-
- MedSwin’s deployable 7B model is trained for reliability rather than raw scale,
- combining large-scale augmentation, supervised fine-tuning, and knowledge distillation.
-
-
-
-
-
A · Data augmentation
-
- Paraphrasing, formatting variants, deduplication, and medical consistency checks expand coverage without semantic drift.
-
-
-
-
B · Supervised fine-tuning
-
- Aligns the student to clinical instruction style, tone control, and structured answers.
-
-
-
-
C · Knowledge distillation
-
- Hard labels expand task coverage; soft labels preserve calibration and uncertainty from a larger instructor.
-
-
-
+
+
+
Retrieval and sufficiency
+
Evidence is selected by clinical utility, not by raw top-K truncation.
+
Each retrieval stage uses a responsive diagram, and the sufficiency simulator animates toward threshold when more evidence is retrieved.
+
+
+
+
+
+
+
+
+
+
Facet sufficiency simulator
Build an evidence bundle
+
-
-
-
+
+
+
+
Critical facets are below acceptance threshold.
-
-
-
-
-
-
-
-
-
- Evaluation & Safety
-
-
- MedSwin evaluates clinical QA systems beyond answer accuracy, focusing on evidence quality,
- guideline compliance, and runtime safety behaviour.
-
-
-
-
-
- Retrieval quality
-
-
- Evidence relevance and coverage under a fixed token budget.
-
-
-
-
- Guideline coverage
-
-
- Presence of actionable recommendations and contraindications.
-
-
-
-
- Faithfulness
-
-
- Final answers remain grounded in cited evidence only.
-
-
-
+
+
+
Multi-agent coordination
+
The MAC layer behaves like a specialist dive team.
+
Agents explore different hypotheses, return claim-level ledgers, and are aggregated through reliability-weighted evidence selection.
+
+
+
+
+
Claim-level ledger
Every claim keeps its source role.
audit artifact
+
+
+
Facet
Role
Polarity
Grade
Trace
+
+
Guideline concordance
Recommendation
supports
CPG
doc · version · section
+
Patient applicability
Lab / comorbidity
qualifies
EMR
encounter · timestamp
+
Safety risk
Contraindication
conflicts
Safety
severity · source
+
Uncertainty
Contradiction pair
preserved
Mixed
adjudication status
+
+
-
-
-
+
Contradictions are not averaged away.High-grade conflicts are preserved until the critic or final synthesiser explicitly adjudicates them.
-
-
-
-
-
-
Team
-
- A multidisciplinary research team building an auditable medical AI system.
-
+
+
+
Evaluation dashboard
+
QA, reranking, and audit metrics are inspected separately.
+
Overlap metrics, semantic similarity, retrieval quality, latency, and audit completeness are kept in separate views so answer similarity is not mistaken for deployment readiness.
+
+
+
+
+
+
+
-
- Swinburne
- Multi-role
+
+
+
-
-
-
-
🎖️
-
Liam
-
Leader
-
-
-
🧪
-
Henry
-
LLM
-
-
-
🔗
-
Hai
-
System
+
+
+
+
Model
ROUGE-L
BERT-F1
Token F1
Uni Prec
Bi Prec
+
+
+
+
+
+
MSAS component families
Auditability is separated from answer similarity.
+
Clinical interpretation boundary
Overlap and semantic metrics are useful signals, but unsupported claims, unsafe omissions, missing provenance, and unresolved contradictions are tracked separately because answer similarity alone is not enough for deployment safety.
-