Back to Medlingo
Responsible AI · Clinical Use Policy

Use cases, limitations,
and honest caveats

Medlingo is a powerful clinical reasoning assistant — but it is not a replacement for a doctor. This page explains exactly where it helps, where to verify, and what it cannot do.

⚕
Important — Read Before Use
Medlingo is a clinical decision support tool, not a diagnostic authority. All outputs — including differential diagnoses, medication suggestions, and clinical summaries — must be verified by a licensed medical professional before acting upon them. The model can and does hallucinate. It may produce plausible-sounding but clinically incorrect information. Your judgment as a physician always takes precedence.
Use Cases

Where Medlingo excels —
and where it needs your oversight

We've categorized every major use case by confidence level. This is our honest assessment — not marketing.

✓ High confidence — Works well
✓ Strong
Differential Diagnosis Generation
Given a clear set of clinical findings, Medlingo reliably generates a ranked differential with reasoning. Particularly strong for common GP presentations.
✓
Chest pain differentials in adults
✓
Respiratory presentations with vitals
✓
GI complaint workup suggestions
✓
Common endocrine pattern recognition
✓ Strong
Patient History Summarization
Consolidating a patient's existing records into a structured, readable summary for rapid orientation at the start of a visit.
✓
Medication reconciliation overview
✓
Allergy and contraindication flagging
✓
Visit history timeline
✓
Lab result trend summaries
✓ Strong
Clinical Question Answering
Answering general clinical questions in the context of the loaded patient — drug dosages, first-line treatments, diagnostic criteria.
✓
First-line treatment for common conditions
✓
Diagnostic criteria recall (e.g. Wells score)
✓
Drug class overviews and MOA
⚠ Proceed with caution — Verify before acting
⚠ Verify
Specific Drug Dosing
Medlingo can suggest dosing ranges, but exact dosing — especially for renally-adjusted or weight-based regimens — must always be cross-checked against a verified formulary.
!
Paediatric dosing calculations
!
Renal/hepatic dose adjustments
!
High-alert medication ranges
⚠ Verify
Rare or Atypical Presentations
Uncommon diagnoses, rare diseases, or highly atypical presentations may be under-represented in training data. The model may anchor on common diagnoses incorrectly.
!
Rare genetic or metabolic disorders
!
Atypical MI presentations in women
!
Overlapping multi-system conditions
⚠ Verify
Lab Value Interpretation
General lab interpretation is reliable. But reference ranges vary by lab, and clinical context (e.g. pregnancy, altitude, chronic disease) changes normal values significantly.
!
Always use your lab's reference range
!
Trending values over time is more reliable
✕ Not intended for these
✕ Not for this
Emergency Triage Decisions
Medlingo is not designed for time-critical emergency decisions. In any emergency setting, follow established triage and resuscitation protocols without reliance on this tool.
✕
STEMI / active cardiac arrest management
✕
Stroke code activation criteria
✕
Trauma primary survey decisions
✕ Not for this
Final Diagnosis Issuance
Medlingo provides differentials and reasoning — it does not issue diagnoses. No output from this system should ever appear on a formal clinical document as a confirmed diagnosis.
✕
Discharge diagnoses
✕
Death certificates or medico-legal documents
✕
Insurance or referral documentation
✕ Not for this
Patient-Facing Communication
Medlingo is a tool for clinicians, not patients. Its outputs are designed for physician interpretation and should not be shared directly with patients as clinical advice.
✕
Direct patient messaging or counselling
✕
Automated patient report generation
AI Limitations

It will hallucinate.
Here's what that means.

Hallucination is not a bug unique to Medlingo — it is an inherent property of all large language models. We are transparent about this because your patients' safety depends on you understanding it.

⚠️
What hallucination looks like
Real patterns observed in clinical LLMs
💊
Fabricated drug interactions
The model may confidently state a drug interaction exists that doesn't, or miss one that does. Always cross-reference a verified interactions database.
🔢
Plausible but wrong numbers
Dosing figures, lab reference ranges, and diagnostic cutoffs may be slightly off — generated to sound correct, not looked up.
📚
Outdated clinical guidelines
Training data has a cutoff date. Guidelines change. Medlingo may reflect older protocols — always verify against current guidelines.
🎭
Confident tone despite uncertainty
LLMs don't naturally express uncertainty well. A confident, well-structured answer is not evidence of correctness. Treat all outputs as hypotheses, not facts.
✅
Why Medlingo is still reliable
Our claim: better than most models in this domain
🎯
Domain-focused training
Medlingo is fine-tuned on clinical data, not general internet text. This reduces but does not eliminate hallucination on medical facts.
🧠
Patient-grounded reasoning
Every query runs against the patient's actual loaded record. The model reasons from real data, not invented context — reducing fabrication risk.
📊
Comparative benchmark performance
In internal testing against general-purpose LLMs on GP case scenarios, Medlingo showed fewer factual errors and more appropriate uncertainty flagging. Independent audit is planned.
🔄
Human-in-the-loop by design
The interface is built to surface reasoning chains, not just conclusions — so you can spot errors before acting. Every output is a starting point for your judgment.
Our honest claim

Medlingo produces more reliable clinical differentials than general-purpose AI assistants in GP settings — because it is built specifically for this task and has access to patient context.

However, no AI system should be trusted without physician verification in a clinical setting. The performance gain is real, but so is the residual error rate.

Comparison

Medlingo vs. general AI
in clinical use

A feature-level comparison of Medlingo against using a general-purpose LLM for the same tasks.

Capability
Medlingo
General LLM
Manual (no AI)
Clinical reasoning
Differential diagnosis generation
✓
~
✗
Patient-aware reasoning
✓
✗
✗
Confidence-ranked differentials
✓
✗
✗
Patient data
Persistent patient history
✓
✗
~
Allergy & drug flag integration
✓
✗
~
Lab result context
✓
✗
~
Safety
Explicit AI disclaimer on every output
✓
✗
✓
Visible reasoning chain
✓
~
✗
✓ Fully supported  ·  ~ Partial / inconsistent  ·  ✗ Not supported
Safe Use Protocols

How to use Medlingo
safely

Four practical protocols every doctor using Medlingo should follow in every session.

Protocol 01
Always verify the differential
Treat every Medlingo differential as a list of hypotheses to test, not a list of confirmed options. Verify the top suggestion against your clinical findings independently.
1
Review the AI's reasoning chain, not just the output
2
Check each differential against your own exam findings
3
Confirm patient allergies before any prescription decision
Protocol 02
Cross-check all dosing externally
Never prescribe a dose based solely on Medlingo's suggestion. Use a verified formulary — BNF, MIMS, or your institution's drug reference — for all prescribing decisions.
1
Use Medlingo for drug class and indication reasoning
2
Cross-reference exact dose with BNF / MIMS
3
Adjust for renal, hepatic, age, and weight as indicated
Protocol 03
Do not use in emergencies
In time-critical situations, follow your institution's emergency protocols and established clinical guidelines. Medlingo is a tool for elective clinical reasoning, not emergency decision-making.
1
Identify emergency presentations immediately
2
Follow ACLS / ALS / institutional protocol
3
Consult Medlingo only after stabilization, if needed
Protocol 04
Maintain clinical documentation
Your clinical notes and decisions remain your responsibility. Medlingo session outputs are for your reasoning support only — document your own conclusions in the patient record.
1
Never copy-paste AI output into official clinical notes
2
Document your own reasoning and decision-making
3
Maintain full accountability for patient care

Full Medical Disclaimer

⚕️
Not a licensed medical device
Medlingo is a Final Year Project prototype and has not been reviewed, certified, or approved by any medical regulatory authority (PMDA, FDA, CE Mark, DRAP, or equivalent). It is not a licensed medical device and must not be used as the sole basis for any clinical decision.
🔬
AI outputs may be incorrect
This system uses a large language model that can produce factually incorrect, outdated, or misleading clinical information even when stated with apparent confidence. This is a known property of all current LLM systems.
👨‍⚕️
Physician judgment is mandatory
Every output from Medlingo — including differentials, summaries, and treatment suggestions — must be reviewed and validated by a licensed medical professional before influencing any patient care decision.
📋
Liability
The Medlingo development team accepts no liability for clinical outcomes resulting from use of this tool. The treating physician bears full responsibility for all diagnostic and treatment decisions made during patient care.
By using Medlingo, you acknowledge that you are a licensed medical professional, that you understand the limitations of AI-generated clinical content, and that you accept sole responsibility for all decisions made in the care of your patients.