Metadata-Version: 2.4
Name: triagesim
Version: 0.2.0.post1
Summary: Multi-speaker spoken dialogue simulation framework for emergency department triage
Author-email: Dipankar Srirag <d.srirag@unsw.edu.au>
License: 
                                         Apache License
                                   Version 2.0, January 2004
                                http://www.apache.org/licenses/
        
           TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
        
           1. Definitions.
        
              "License" shall mean the terms and conditions for use, reproduction,
              and distribution as defined by Sections 1 through 9 of this document.
        
              "Licensor" shall mean the copyright owner or entity authorized by
              the copyright owner that is granting the License.
        
              "Legal Entity" shall mean the union of the acting entity and all
              other entities that control, are controlled by, or are under common
              control with that entity. For the purposes of this definition,
              "control" means (i) the power, direct or indirect, to cause the
              direction or management of such entity, whether by contract or
              otherwise, or (ii) ownership of fifty percent (50%) or more of the
              outstanding shares, or (iii) beneficial ownership of such entity.
        
              "You" (or "Your") shall mean an individual or Legal Entity
              exercising permissions granted by this License.
        
              "Source" form shall mean the preferred form for making modifications,
              including but not limited to software source code, documentation
              source, and configuration files.
        
              "Object" form shall mean any form resulting from mechanical
              transformation or translation of a Source form, including but
              not limited to compiled object code, generated documentation,
              and conversions to other media types.
        
              "Work" shall mean the work of authorship, whether in Source or
              Object form, made available under the License, as indicated by a
              copyright notice that is included in or attached to the work
              (an example is provided in the Appendix below).
        
              "Derivative Works" shall mean any work, whether in Source or Object
              form, that is based on (or derived from) the Work and for which the
              editorial revisions, annotations, elaborations, or other modifications
              represent, as a whole, an original work of authorship. For the purposes
              of this License, Derivative Works shall not include works that remain
              separable from, or merely link (or bind by name) to the interfaces of,
              the Work and Derivative Works thereof.
        
              "Contribution" shall mean any work of authorship, including
              the original version of the Work and any modifications or additions
              to that Work or Derivative Works thereof, that is intentionally
              submitted to Licensor for inclusion in the Work by the copyright owner
              or by an individual or Legal Entity authorized to submit on behalf of
              the copyright owner. For the purposes of this definition, "submitted"
              means any form of electronic, verbal, or written communication sent
              to the Licensor or its representatives, including but not limited to
              communication on electronic mailing lists, source code control systems,
              and issue tracking systems that are managed by, or on behalf of, the
              Licensor for the purpose of discussing and improving the Work, but
              excluding communication that is conspicuously marked or otherwise
              designated in writing by the copyright owner as "Not a Contribution."
        
              "Contributor" shall mean Licensor and any individual or Legal Entity
              on behalf of whom a Contribution has been received by Licensor and
              subsequently incorporated within the Work.
        
           2. Grant of Copyright License. Subject to the terms and conditions of
              this License, each Contributor hereby grants to You a perpetual,
              worldwide, non-exclusive, no-charge, royalty-free, irrevocable
              copyright license to reproduce, prepare Derivative Works of,
              publicly display, publicly perform, sublicense, and distribute the
              Work and such Derivative Works in Source or Object form.
        
           3. Grant of Patent License. Subject to the terms and conditions of
              this License, each Contributor hereby grants to You a perpetual,
              worldwide, non-exclusive, no-charge, royalty-free, irrevocable
              (except as stated in this section) patent license to make, have made,
              use, offer to sell, sell, import, and otherwise transfer the Work,
              where such license applies only to those patent claims licensable
              by such Contributor that are necessarily infringed by their
              Contribution(s) alone or by combination of their Contribution(s)
              with the Work to which such Contribution(s) was submitted. If You
              institute patent litigation against any entity (including a
              cross-claim or counterclaim in a lawsuit) alleging that the Work
              or a Contribution incorporated within the Work constitutes direct
              or contributory patent infringement, then any patent licenses
              granted to You under this License for that Work shall terminate
              as of the date such litigation is filed.
        
           4. Redistribution. You may reproduce and distribute copies of the
              Work or Derivative Works thereof in any medium, with or without
              modifications, and in Source or Object form, provided that You
              meet the following conditions:
        
              (a) You must give any other recipients of the Work or
                  Derivative Works a copy of this License; and
        
              (b) You must cause any modified files to carry prominent notices
                  stating that You changed the files; and
        
              (c) You must retain, in the Source form of any Derivative Works
                  that You distribute, all copyright, patent, trademark, and
                  attribution notices from the Source form of the Work,
                  excluding those notices that do not pertain to any part of
                  the Derivative Works; and
        
              (d) If the Work includes a "NOTICE" text file as part of its
                  distribution, then any Derivative Works that You distribute must
                  include a readable copy of the attribution notices contained
                  within such NOTICE file, excluding those notices that do not
                  pertain to any part of the Derivative Works, in at least one
                  of the following places: within a NOTICE text file distributed
                  as part of the Derivative Works; within the Source form or
                  documentation, if provided along with the Derivative Works; or,
                  within a display generated by the Derivative Works, if and
                  wherever such third-party notices normally appear. The contents
                  of the NOTICE file are for informational purposes only and
                  do not modify the License. You may add Your own attribution
                  notices within Derivative Works that You distribute, alongside
                  or as an addendum to the NOTICE text from the Work, provided
                  that such additional attribution notices cannot be construed
                  as modifying the License.
        
              You may add Your own copyright statement to Your modifications and
              may provide additional or different license terms and conditions
              for use, reproduction, or distribution of Your modifications, or
              for any such Derivative Works as a whole, provided Your use,
              reproduction, and distribution of the Work otherwise complies with
              the conditions stated in this License.
        
           5. Submission of Contributions. Unless You explicitly state otherwise,
              any Contribution intentionally submitted for inclusion in the Work
              by You to the Licensor shall be under the terms and conditions of
              this License, without any additional terms or conditions.
              Notwithstanding the above, nothing herein shall supersede or modify
              the terms of any separate license agreement you may have executed
              with Licensor regarding such Contributions.
        
           6. Trademarks. This License does not grant permission to use the trade
              names, trademarks, service marks, or product names of the Licensor,
              except as required for reasonable and customary use in describing the
              origin of the Work and reproducing the content of the NOTICE file.
        
           7. Disclaimer of Warranty. Unless required by applicable law or
              agreed to in writing, Licensor provides the Work (and each
              Contributor provides its Contributions) on an "AS IS" BASIS,
              WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
              implied, including, without limitation, any warranties or conditions
              of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
              PARTICULAR PURPOSE. You are solely responsible for determining the
              appropriateness of using or redistributing the Work and assume any
              risks associated with Your exercise of permissions under this License.
        
           8. Limitation of Liability. In no event and under no legal theory,
              whether in tort (including negligence), contract, or otherwise,
              unless required by applicable law (such as deliberate and grossly
              negligent acts) or agreed to in writing, shall any Contributor be
              liable to You for damages, including any direct, indirect, special,
              incidental, or consequential damages of any character arising as a
              result of this License or out of the use or inability to use the
              Work (including but not limited to damages for loss of goodwill,
              work stoppage, computer failure or malfunction, or any and all
              other commercial damages or losses), even if such Contributor
              has been advised of the possibility of such damages.
        
           9. Accepting Warranty or Additional Liability. While redistributing
              the Work or Derivative Works thereof, You may choose to offer,
              and charge a fee for, acceptance of support, warranty, indemnity,
              or other liability obligations and/or rights consistent with this
              License. However, in accepting such obligations, You may act only
              on Your own behalf and on Your sole responsibility, not on behalf
              of any other Contributor, and only if You agree to indemnify,
              defend, and hold each Contributor harmless for any liability
              incurred by, or claims asserted against, such Contributor by reason
              of your accepting any such warranty or additional liability.
        
           END OF TERMS AND CONDITIONS
        
           APPENDIX: How to apply the Apache License to your work.
        
              To apply the Apache License to your work, attach the following
              boilerplate notice, with the fields enclosed by brackets "[]"
              replaced with your own identifying information. (Don't include
              the brackets!)  The text should be enclosed in the appropriate
              comment syntax for the file format. We also recommend that a
              file or class name and description of purpose be included on the
              same "printed page" as the copyright notice for easier
              identification within third-party archives.
        
           Copyright [2025] [Dipankar Srirag]
        
           Licensed under the Apache License, Version 2.0 (the "License");
           you may not use this file except in compliance with the License.
           You may obtain a copy of the License at
        
               http://www.apache.org/licenses/LICENSE-2.0
        
           Unless required by applicable law or agreed to in writing, software
           distributed under the License is distributed on an "AS IS" BASIS,
           WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
           See the License for the specific language governing permissions and
           limitations under the License.
Project-URL: Homepage, https://github.com/dipankarsrirag/triage-sim
Project-URL: Repository, https://github.com/dipankarsrirag/triage-sim
Project-URL: Issues, https://github.com/dipankarsrirag/triage-sim/issues
Keywords: triage,simulation,dialogue,speech,NLP,healthcare,emergency medicine,spoken dialogue
Classifier: Development Status :: 3 - Alpha
Classifier: Intended Audience :: Science/Research
Classifier: License :: OSI Approved :: MIT License
Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
Classifier: Programming Language :: Python :: 3
Classifier: Programming Language :: Python :: 3.11
Classifier: Programming Language :: Python :: 3.12
Classifier: Programming Language :: Python :: 3.13
Requires-Python: >=3.11
Description-Content-Type: text/markdown
License-File: LICENSE
Requires-Dist: pydantic>=2.9
Requires-Dist: openai>=1.40
Requires-Dist: python-dotenv>=1.0
Requires-Dist: PyYAML>=6.0
Provides-Extra: vllm
Requires-Dist: vllm>=0.11; extra == "vllm"
Provides-Extra: audio
Requires-Dist: torch>=2.0; extra == "audio"
Requires-Dist: TTS>=0.22; extra == "audio"
Provides-Extra: dev
Requires-Dist: pytest>=8.0; extra == "dev"
Dynamic: license-file

# TriageSim

**TriageSim** is an open-source Python framework for generating synthetic, multi-speaker spoken dialogues for emergency department (ED) triage. It produces paired structured EHR → dialogue data grounded in real clinical vignettes, enabling controlled evaluation of speech and language systems in healthcare.

---

## Features

- **Nurse ↔ patient dialogue simulation** with LLM agents served by **vLLM** (local, in-process) or **OpenRouter** (hosted models: Claude, Gemini, GPT, ...)
- **Batched generation**: many episodes run side by side and every round of LLM calls goes to vLLM as one batch
- **Schema-constrained outputs**: agents answer in JSON that is enforced during decoding, including which actions and vitals the nurse may pick at each step
- **Two triage algorithms**: ESI (Emergency Severity Index) and ATS (Australasian Triage Scale)
- **Research-grounded personas**: combinatorial patient and nurse personas built from ED-research traits with behavioural anchors (health literacy, symptom reporting, disclosure, questioning style, ...)
- **Structured run artifacts**: dialogue, vitals checked, red flags logged, per-step triage reasoning and the nurse's triage record (its beliefs, kept by the dialogue master)
- **Reproducible and resumable** runs to JSONL, straight from MIMIC-IV-ED `triage.csv` or your own vignettes
- **Optional audio rendering**: XTTS-v2 voice cloning for multi-speaker TTS synthesis

---

## Installation

```bash
pip install triagesim            # OpenRouter backend
pip install "triagesim[vllm]"    # + local models with vLLM (needs a CUDA GPU)
pip install "triagesim[audio]"   # + XTTS-v2 speech rendering
```

Or the latest development version from GitHub:
```bash
pip install "git+https://github.com/dipankarsrirag/triage-sim.git"
```

For OpenRouter, set your [API key](https://openrouter.ai/) in the environment or in a `.env` file:

```bash
export OPENROUTER_API_KEY=sk-or-...
```

---

## Quick Start (command line)

Models are given as `vllm:<local path or Hugging Face id>` or `openrouter:<model id>`:

```bash
triagesim --model vllm:google/gemma-4-E4B-it \
    --cases examples/cases.jsonl \
    --nurse-personas examples/personas/nurse.yaml \
    --patient-personas examples/personas/patient.yaml \
    --out outputs/gemma.jsonl \
    --vllm-args '{"max_model_len": 16384}'
```

```bash
triagesim --model openrouter:anthropic/claude-sonnet-4.5 --algorithm ats --episodes-per-case 3 \
    --cases examples/cases.jsonl \
    --nurse-personas examples/personas/nurse.yaml \
    --patient-personas examples/personas/patient.yaml \
    --out outputs/claude.jsonl
```

Each role can use its own model (`--nurse-model`, `--patient-model`, `--judge-model`),
each either vLLM or OpenRouter, e.g. a hosted nurse against a local patient simulator. Different vLLM
models share one GPU and split its memory evenly unless `--vllm-args` sets `gpu_memory_utilization`.

`--judge-model` turns on the dialogue master, which screens cases, writes the patient's script and
checks every spoken line (`--max-regenerations`, default 1).

Episodes are appended to `--out` as they finish; re-running the same command skips episodes already
there, so an interrupted run just resumes. Ctrl-C or SIGTERM stops a run at once: finished episodes
are kept, and the ones in flight run again on the next run. Episodes whose LLM calls keep failing go to
`<out>.errors.jsonl` and are retried on the next run. At the end the command prints summary metrics.
`triagesim --help` lists the other options (`--max-turns`, `--sampling`, `--concurrency`, `--seed`, ...);
arguments can also come from a file, one per line: `triagesim @run.args`.

---

## Python API

```python
from triagesim import (
    VLLM, OpenRouter, NursePersona, PatientPersona,
    load_cases, load_personas, simulate, summarize,
)

llm = VLLM("google/gemma-4-E4B-it", max_model_len=16384)   # or OpenRouter("anthropic/claude-sonnet-4.5")

episodes = list(simulate(
    load_cases("examples/cases.jsonl"),
    nurse_llm=llm,
    patient_llm=llm,
    nurse_personas=load_personas("examples/personas/nurse.yaml", NursePersona),
    patient_personas=load_personas("examples/personas/patient.yaml", PatientPersona),
    max_turns=12,
    judge_llm=llm,               # optional dialogue master
    sampling={"nurse": {"temperature": 0.7}},
))
print(summarize(episodes))
```

`simulate` yields one dict per episode as it finishes, so large runs can be streamed to disk.
Keyword arguments to `VLLM` go to `vllm.LLM` (plus `chat_template_kwargs`, e.g.
`{"enable_thinking": False}`); sampling parameters per role (`nurse`, `patient`, `judge`, `script`)
override the model's `generation_config` defaults.

---

## Cases

A case is the ground truth for one presentation. The nurse never sees it: it learns only what the
patient says, what is visible at the triage desk (gender, age group, arrival mode) and the vital signs
it asks for. The patient knows its complaint, pain, and how unwell it is (the acuity, in plain words).

```json
{"case_id": "syncope-01", "chief_complaint": "Syncope", "vitals": {"temperature": 99.1, "heartrate": 112, "resprate": 26, "o2sat": 91, "sbp": 98}, "acuity": 2, "acuity_scale": "esi", "pain": 7, "gender": "male", "arrival_transport": "ambulance"}
```

`load_cases` reads `.jsonl`, `.json` or `.csv`. Vitals may also be flat columns, so MIMIC-IV-ED
`triage.csv` loads as is (`stay_id` becomes the case id, `chiefcomplaint` the chief complaint, and
blank values become "not recorded"). Optional fields: `gender` (patient personas are matched to it),
`acuity_scale` (`esi`/`ats`: the scale the nurse triages with unless `--algorithm` is given),
`diagnoses` and `medications` (e.g. MIMIC-IV-ED ICD titles and medrecon; they ground the patient's
script). Any other field is kept in the artifacts. `scripts/build_mimic_cases.py` builds cases from the
MIMIC-IV-ED tables.

### The dialogue master

With `--judge-model`, the dialogue master runs each episode around the two agents:

1. **In-situ check.** It reasons from the case whether the patient would have a triage conversation
   at the desk at all (not, for example, when they go straight to resuscitation). If not, the episode
   is recorded with status `"skipped"` and the master's reasoning. Cases without a ground-truth acuity
   are always skipped.
2. **Patient script.** It writes a standardized-patient script from the case and the patient persona:
   story, onset, quality, associated symptoms, pertinent negatives, history, medications,
   allergies. The patient answers from it.
3. **Line checks.** It verifies every spoken line (see below).

### Building an evaluation pool

Run many simulations, over many model pairings and personas, then pool them and keep the episodes
whose nurse reached the ground-truth level (with optional quality filters):

```bash
python scripts/select_episodes.py runs/*.jsonl --out pool.jsonl --no-flags --nurse-ended
```

It reports how many cases the pool covers, by acuity and by model pairing.

---

## Personas

Personas are combinations of traits grounded in emergency-department research. Each trait has a
definition, levels anchored in observable behaviour, and sources; the registry in
`triagesim/personas.py` is the single source of truth, and `docs/persona-traits-review.md` holds the
evidence and the design decisions.

| Patient trait | Levels |
|---|---|
| `age_group` | adult, older |
| `english_proficiency` | fluent, functional, limited |
| `health_literacy` | limited, marginal, adequate |
| `recall_reliability` | reliable, patchy, unreliable |
| `symptom_reporting` | minimising, faithful, amplifying |
| `distress` | calm, anxious, highly_distressed |
| `disclosure` | open, selective, guarded |
| `demeanour` | cooperative, impatient, hostile |
| `verbosity` | terse, typical, expansive |
| `focus` | on_target, occasional_tangents, frequent_tangents |
| `disfluency` | low, moderate, high |

| Nurse trait (interview style only) | Levels |
|---|---|
| `expertise` | novice, competent, expert |
| `algorithm_reliance` | protocol_driven, blended, intuition_led |
| `risk_attitude` | risk_averse, balanced, risk_accepting |
| `self_report_credence` | accepting, corroborating, discounting |
| `questioning_style` | open_facilitative, mixed, closed_directive |
| `emotional_responsiveness` | task_focused, acknowledging, empathic |
| `workload` | quiet, busy, overloaded |

The models see each trait with its behavioural anchor. A patient's gender comes from the case, and
its cognitive state and intoxication are set by the dialogue master from the case (they are triage
signs, not personality). Nurse traits shape which questions are asked and how, never the level
assigned. `ethnicity` and `instruction` (voice direction for TTS) may be stored on a persona but are
never shown to the models.

Pools are generated combinatorially: every combination of levels except the two pairs that
contradict each other by definition (`personas.INCOMPATIBLE`; the review of all trait pairs is in
`docs/persona-pair-review.md`). That gives 104,976 patient and 3,888 nurse personas:

```bash
python scripts/generate_personas.py --out-dir data/personas
```

`load_personas` reads these JSONL pools, or YAML lists such as `examples/personas/`.

---

## How an Episode Runs

Each turn the nurse acts until it speaks: it may first check one vital sign it does not know yet
(`check_vital`, the value comes from the case; what it says while taking one, e.g. "Let me check
your blood pressure.", enters the dialogue if longer than two words), so vitals are spread through
the conversation without using up questions, and log red flags once
(`log_red_flag`), then either asks the patient something (`utterance`) or ends triage (`end`). Every
nurse step also gives its rationale, then its current triage estimate and confidence, and is told how
many questions it has left (`max_turns`, default 12); only questions use them up. The patient then
replies once.
The nurse may end only after `min_questions` questions (default 3) and, with a dialogue master,
once every field of its triage record holds an answer ("none" and "declined to say" count) or its
questions run out; the prompt lists what a triage interview usually covers. After its last question is answered it gets one more
step without questions: it may check any vitals it still needs and log red flags, then ends with its final level
(`end_reason` is then `"max_turns"`). When it ends it says a brief closing line to the patient
(never the triage level).

With a dialogue master, every nurse and patient line is checked before it enters the dialogue for
faithfulness (to the case and script, or to what the nurse has actually learned), information gain,
persona adherence, real-world plausibility and the absence of names; a nurse line must also ask for
one thing only (no two questions joined by "and"; a list of related symptoms or an either/or choice
counts as one question). A rejected line goes back to
the speaker with the master's reason; if the new line still fails, the master writes the line itself
(`verification.edited`), keeping the speaker's intent and persona. The rejected drafts are kept.

The nurse's beliefs are its **triage record**: chief complaint, onset and course, pain, associated
symptoms, relevant history, medications, allergies, and the vital signs taken. With a dialogue master,
after every patient reply the master reads the exchange (the nurse's question and the reply, words
alone) and writes what it conveyed into the record; vital signs are added as they are taken. The
nurse sees the record, with its gaps, at every step, and the final record is saved with the episode.

Every patient line also records what information passed, from three sides, under `information`:
`disclosed` (the patient's own one-sentence account of what it revealed), `understood` (the nurse's
one-sentence reading of it, written in its next step before it acts) and, with a dialogue master,
`conveyed` (the master's one-sentence reading of the exchange's words alone, without the case) and
`recorded` (what it added to the triage record).

---

## Run Artifact

Each episode is a dict (one JSONL line):

| Key | Description |
|---|---|
| `episode_id` | `<case_id>-<k>` |
| `status`, `error` | `"ok"`, or `"error"` with the reason (dialogue kept up to the failure) |
| `case` | The ground-truth case |
| `algorithm`, `models` | Triage algorithm and the model used for each role |
| `nurse_persona`, `patient_persona` | Personas drawn for this episode |
| `patient_script` | The script drawn for this episode, if any |
| `end_reason` | `"nurse_end"` or `"max_turns"` |
| `final_triage`, `num_turns` | The nurse's triage level at its last step; number of patient replies |
| `transcript` | The output transcript: spoken nurse and patient lines only (`speaker`, `text`); lines kept for the agents alone, such as announcing a vital that could not be measured, are left out |
| `history` | Dialogue and events: utterances (patient lines with `information`, and `verification` when a judge is used), vitals, triage end |
| `trace` | Every nurse step: action, vital, utterance, triage, confidence, red flags, explanation |
| `red_flags` | Red flags logged by the nurse (deduplicated) |
| `record` | The nurse's final triage record: the fields the dialogue master filled in, and `vitals` taken |

---

## Evaluation Metrics

```python
from triagesim import score, summarize

score(episode)       # correctness, absolute error, over/under-triage, first correct turn, flagged lines, ...
summarize(episodes)  # accuracy, mean absolute error, over/under-triage, flag rates, ...
```

---

## Running on a PBS Cluster

`jobs/simulate.pbs` runs a simulation on one GPU (vLLM start-up takes a few minutes, so batch many
episodes per job), `jobs/evaluate.pbs` measures triage accuracy on the seed datasets in `data/`
(`scripts/evaluate.py`), and `jobs/smoke_test.pbs` checks both backends end to end on a small model.
They are set up for UNSW Katana; adjust the module and scratch paths for your cluster.

---

## Citation

If you use TriageSim in your research, please cite:

```bibtex
@inproceedings{srirag26_interspeech,
  title     = {{TriageSim: A Conversational Emergency Triage Simulation Framework from Structured Electronic Health Records}},
  author    = {Dipankar Srirag and Quoc Dung Nguyen and Aditya Joshi and Padmanesan Narasimhan and Salil Kanhere},
  year      = {2026},
  booktitle = {{Interspeech 2026}},
  pages     = {1997--2001},
  doi       = {10.21437/Interspeech.2026-819},
  issn      = {2958-1796},
}
```

---

## License

MIT © Dipankar Srirag
