# pritamdeka/sbert2s1 architecture

> Architecture diagram of the GitHub repository pritamdeka/sbert2s1, generated by GitDiagram from its file tree, README and a sample of source files (last updated 2026-10-04). It is an AI-made overview: check details against the source.

- Interactive diagram: https://gitdiagram.com/pritamdeka/sbert2s1
- Repository: https://github.com/pritamdeka/sbert2s1

## Overview

sbert2s1 packages a public typed-decision inference model and the research pipeline used to build, train, calibrate, evaluate, and analyse it. The user workflow is to load an exported encoder model, submit text plus typed questions, and receive calibrated option probabilities. Research additionally builds biomedical datasets, runs encoder and generative-LLM experiments, and produces paper analyses. The graph follows documented capabilities and indexed references; unsampled implementation details are not treated as verified calls.

Main entry points: [`research/s1/train.py`](https://github.com/pritamdeka/sbert2s1/blob/main/research/s1/train.py) (Training Orchestrator), [`research/s1/evaluate.py`](https://github.com/pritamdeka/sbert2s1/blob/main/research/s1/evaluate.py) (Prediction Evaluation), [`research/s1/data.py`](https://github.com/pritamdeka/sbert2s1/blob/main/research/s1/data.py) (Tokenized Batches), [`research/s1/llm_baseline.py`](https://github.com/pritamdeka/sbert2s1/blob/main/research/s1/llm_baseline.py) (Generative Baseline)

## Diagram (Mermaid)

```mermaid
flowchart TD

subgraph group_public["Public Inference"]
  node_request["Typed Request"]
  node_predictor["S1 Predictor<br/>[sbert2s1.py]"]
end

subgraph group_data["Biomedical Data"]
  node_public_builder["Public Task Builder<br/>[build_public.py]"]
  node_clinical_builder["Clinical Builder<br/>[build_clinical.py]"]
  node_medline_builder["MEDLINE Builder<br/>[build_medline.py]"]
  node_medline_hydrate["Abstract Hydration<br/>[hydrate_medline.py]"]
  node_schema["Decision Schema<br/>[schema.py]"]
  node_prepared["Prepared Splits"]
end

subgraph group_runtime["Research Runtime"]
  node_token_data["Tokenized Batches<br/>[data.py]"]
  node_model["Decision Architectures<br/>[model.py]"]
  node_model_io["Model Resolution<br/>[models_io.py]"]
  node_objectives["Training Objectives<br/>[objectives.py]"]
  node_trainer["Training Orchestrator<br/>[train.py]"]
  node_calibrator["Temperature Fitting<br/>[calibrate.py]"]
  node_lookup["Temperature Lookup"]
  node_evaluator["Prediction Evaluation<br/>[evaluate.py]"]
  node_metrics["Decision Metrics<br/>[metrics.py]"]
end

subgraph group_studies["Experiments And Analysis"]
  node_runner["Experiment Runner<br/>[experiments.py]"]
  node_llm["Generative Baseline<br/>[llm_baseline.py]"]
  node_audit["Data Audit<br/>[audit_data.py]"]
  node_summaries["Run Summaries<br/>[summarize.py]"]
  node_paper_analysis["Paper Analysis<br/>[analysis_common.py]"]
  node_cascade["Confidence Cascade<br/>[cascade_v2.py]"]
  node_paper_tables["Results Tables<br/>[results_tables.py]"]
end

node_user(("Caller"))
node_hub["Model Hub"]

node_user -->|"submits"| node_request
node_request -->|"predicts"| node_predictor
node_predictor -->|"returns probabilities"| node_user
node_hub -.->|"provides model"| node_predictor
node_public_builder -->|"formats records"| node_schema
node_clinical_builder -->|"formats records"| node_schema
node_medline_builder -->|"formats records"| node_schema
node_medline_hydrate -.->|"hydrates abstracts"| node_medline_builder
node_public_builder -->|"writes splits"| node_prepared
node_clinical_builder -.->|"writes splits"| node_prepared
node_medline_builder -.->|"writes splits"| node_prepared
node_medline_hydrate -.->|"adds abstracts"| node_prepared
node_prepared -->|"tokenizes"| node_token_data
node_token_data -->|"uses record format"| node_schema
node_token_data -->|"resolves tokenizer"| node_model_io
node_trainer -->|"assembles batches"| node_token_data
node_trainer -->|"builds and trains"| node_model
node_trainer -->|"optimizes with"| node_objectives
node_trainer -->|"fits temperatures"| node_calibrator
node_trainer -->|"evaluates model"| node_evaluator
node_trainer -->|"resolves encoder"| node_model_io
node_calibrator -->|"exports lookup"| node_lookup
node_evaluator -->|"collates batches"| node_token_data
node_evaluator -->|"uses calibration"| node_calibrator
node_evaluator -->|"computes metrics"| node_metrics
node_runner -.->|"launches runs"| node_trainer
node_runner -.->|"launches baseline"| node_llm
node_llm -->|"resolves model"| node_model_io
node_llm -->|"fits temperatures"| node_calibrator
node_llm -->|"reuses evaluation"| node_evaluator
node_audit -.->|"supplies audit data"| node_paper_analysis
node_paper_analysis -->|"applies temperatures"| node_lookup
node_cascade -->|"reads predictions"| node_paper_analysis
node_paper_tables -.->|"summarizes results"| node_summaries

click node_predictor "https://github.com/pritamdeka/sbert2s1/blob/main/inference/sbert2s1.py"
click node_public_builder "https://github.com/pritamdeka/sbert2s1/blob/main/research/data_build/build_public.py"
click node_clinical_builder "https://github.com/pritamdeka/sbert2s1/blob/main/research/data_build/build_clinical.py"
click node_medline_builder "https://github.com/pritamdeka/sbert2s1/blob/main/research/data_build/build_medline.py"
click node_medline_hydrate "https://github.com/pritamdeka/sbert2s1/blob/main/research/data_build/hydrate_medline.py"
click node_schema "https://github.com/pritamdeka/sbert2s1/blob/main/research/s1/schema.py"
click node_token_data "https://github.com/pritamdeka/sbert2s1/blob/main/research/s1/data.py"
click node_model "https://github.com/pritamdeka/sbert2s1/blob/main/research/s1/model.py"
click node_model_io "https://github.com/pritamdeka/sbert2s1/blob/main/research/s1/models_io.py"
click node_objectives "https://github.com/pritamdeka/sbert2s1/blob/main/research/s1/objectives.py"
click node_trainer "https://github.com/pritamdeka/sbert2s1/blob/main/research/s1/train.py"
click node_calibrator "https://github.com/pritamdeka/sbert2s1/blob/main/research/s1/calibrate.py"
click node_lookup "https://github.com/pritamdeka/sbert2s1/blob/main/research/s1/temperature_lookup.py"
click node_evaluator "https://github.com/pritamdeka/sbert2s1/blob/main/research/s1/evaluate.py"
click node_metrics "https://github.com/pritamdeka/sbert2s1/blob/main/research/s1/metrics.py"
click node_runner "https://github.com/pritamdeka/sbert2s1/blob/main/research/experiments.py"
click node_llm "https://github.com/pritamdeka/sbert2s1/blob/main/research/s1/llm_baseline.py"
click node_audit "https://github.com/pritamdeka/sbert2s1/blob/main/research/analysis/audit_data.py"
click node_summaries "https://github.com/pritamdeka/sbert2s1/blob/main/research/analysis/summarize.py"
click node_paper_analysis "https://github.com/pritamdeka/sbert2s1/blob/main/research/paper/scripts/analysis_common.py"
click node_cascade "https://github.com/pritamdeka/sbert2s1/blob/main/research/paper/scripts/cascade_v2.py"
click node_paper_tables "https://github.com/pritamdeka/sbert2s1/blob/main/research/paper/scripts/results_tables.py"

classDef toneNeutral fill:#f8fafc,stroke:#334155,stroke-width:1.5px,color:#0f172a
classDef toneBlue fill:#dbeafe,stroke:#2563eb,stroke-width:1.5px,color:#172554
classDef toneAmber fill:#fef3c7,stroke:#d97706,stroke-width:1.5px,color:#78350f
classDef toneMint fill:#dcfce7,stroke:#16a34a,stroke-width:1.5px,color:#14532d
classDef toneRose fill:#ffe4e6,stroke:#e11d48,stroke-width:1.5px,color:#881337
classDef toneIndigo fill:#e0e7ff,stroke:#4f46e5,stroke-width:1.5px,color:#312e81
classDef toneTeal fill:#ccfbf1,stroke:#0f766e,stroke-width:1.5px,color:#134e4a
class node_request,node_predictor toneBlue
class node_public_builder,node_clinical_builder,node_medline_builder,node_medline_hydrate,node_schema,node_prepared toneAmber
class node_token_data,node_model,node_model_io,node_objectives,node_trainer,node_calibrator,node_lookup,node_evaluator,node_metrics toneMint
class node_runner,node_llm,node_audit,node_summaries,node_paper_analysis,node_cascade,node_paper_tables toneRose
class node_user,node_hub toneIndigo
```

## Components

### Public Inference

- **Typed Request**
- [S1 Predictor](https://github.com/pritamdeka/sbert2s1/blob/main/inference/sbert2s1.py): `inference/sbert2s1.py`

### Biomedical Data

- [Public Task Builder](https://github.com/pritamdeka/sbert2s1/blob/main/research/data_build/build_public.py): `research/data_build/build_public.py`
- [Clinical Builder](https://github.com/pritamdeka/sbert2s1/blob/main/research/data_build/build_clinical.py): `research/data_build/build_clinical.py`
- [MEDLINE Builder](https://github.com/pritamdeka/sbert2s1/blob/main/research/data_build/build_medline.py): `research/data_build/build_medline.py`
- [Abstract Hydration](https://github.com/pritamdeka/sbert2s1/blob/main/research/data_build/hydrate_medline.py): `research/data_build/hydrate_medline.py`
- [Decision Schema](https://github.com/pritamdeka/sbert2s1/blob/main/research/s1/schema.py): `research/s1/schema.py`
- **Prepared Splits**

### Research Runtime

- [Tokenized Batches](https://github.com/pritamdeka/sbert2s1/blob/main/research/s1/data.py): `research/s1/data.py`
- [Decision Architectures](https://github.com/pritamdeka/sbert2s1/blob/main/research/s1/model.py): `research/s1/model.py`
- [Model Resolution](https://github.com/pritamdeka/sbert2s1/blob/main/research/s1/models_io.py): `research/s1/models_io.py`
- [Training Objectives](https://github.com/pritamdeka/sbert2s1/blob/main/research/s1/objectives.py): `research/s1/objectives.py`
- [Training Orchestrator](https://github.com/pritamdeka/sbert2s1/blob/main/research/s1/train.py): `research/s1/train.py`
- [Temperature Fitting](https://github.com/pritamdeka/sbert2s1/blob/main/research/s1/calibrate.py): `research/s1/calibrate.py`
- [Temperature Lookup](https://github.com/pritamdeka/sbert2s1/blob/main/research/s1/temperature_lookup.py): `research/s1/temperature_lookup.py`
- [Prediction Evaluation](https://github.com/pritamdeka/sbert2s1/blob/main/research/s1/evaluate.py): `research/s1/evaluate.py`
- [Decision Metrics](https://github.com/pritamdeka/sbert2s1/blob/main/research/s1/metrics.py): `research/s1/metrics.py`

### Experiments And Analysis

- [Experiment Runner](https://github.com/pritamdeka/sbert2s1/blob/main/research/experiments.py): `research/experiments.py`
- [Generative Baseline](https://github.com/pritamdeka/sbert2s1/blob/main/research/s1/llm_baseline.py): `research/s1/llm_baseline.py`
- [Data Audit](https://github.com/pritamdeka/sbert2s1/blob/main/research/analysis/audit_data.py): `research/analysis/audit_data.py`
- [Run Summaries](https://github.com/pritamdeka/sbert2s1/blob/main/research/analysis/summarize.py): `research/analysis/summarize.py`
- [Paper Analysis](https://github.com/pritamdeka/sbert2s1/blob/main/research/paper/scripts/analysis_common.py): `research/paper/scripts/analysis_common.py`
- [Confidence Cascade](https://github.com/pritamdeka/sbert2s1/blob/main/research/paper/scripts/cascade_v2.py): `research/paper/scripts/cascade_v2.py`
- [Results Tables](https://github.com/pritamdeka/sbert2s1/blob/main/research/paper/scripts/results_tables.py): `research/paper/scripts/results_tables.py`

### Other components

- **Caller**
- **Model Hub**

## Connections

- Caller → Typed Request: submits (evidence: `README.md`)
- Typed Request → S1 Predictor: predicts (evidence: `README.md`)
- S1 Predictor → Caller: returns probabilities (evidence: `README.md`)
- Model Hub → S1 Predictor: provides model (evidence: `README.md`)
- Public Task Builder → Decision Schema: formats records (evidence: `research/data_build/build_public.py`)
- Clinical Builder → Decision Schema: formats records (evidence: `research/data_build/build_clinical.py`)
- MEDLINE Builder → Decision Schema: formats records (evidence: `research/data_build/build_medline.py`)
- Abstract Hydration → MEDLINE Builder: hydrates abstracts (evidence: `README.md`)
- Public Task Builder → Prepared Splits: writes splits (evidence: `research/data_build/build_public.py`)
- Clinical Builder → Prepared Splits: writes splits (evidence: `README.md`)
- MEDLINE Builder → Prepared Splits: writes splits (evidence: `README.md`)
- Abstract Hydration → Prepared Splits: adds abstracts (evidence: `README.md`)
- Prepared Splits → Tokenized Batches: tokenizes (evidence: `research/s1/data.py`)
- Tokenized Batches → Decision Schema: uses record format (evidence: `research/s1/data.py`)
- Tokenized Batches → Model Resolution: resolves tokenizer (evidence: `research/s1/data.py`)
- Training Orchestrator → Tokenized Batches: assembles batches (evidence: `research/s1/train.py`)
- Training Orchestrator → Decision Architectures: builds and trains (evidence: `research/s1/train.py`)
- Training Orchestrator → Training Objectives: optimizes with (evidence: `research/s1/train.py`)
- Training Orchestrator → Temperature Fitting: fits temperatures (evidence: `research/s1/train.py`)
- Training Orchestrator → Prediction Evaluation: evaluates model (evidence: `research/s1/train.py`)
- Training Orchestrator → Model Resolution: resolves encoder (evidence: `research/s1/train.py`)
- Temperature Fitting → Temperature Lookup: exports lookup (evidence: `research/s1/calibrate.py`)
- Prediction Evaluation → Tokenized Batches: collates batches (evidence: `research/s1/evaluate.py`)
- Prediction Evaluation → Temperature Fitting: uses calibration (evidence: `research/s1/evaluate.py`)
- Prediction Evaluation → Decision Metrics: computes metrics (evidence: `research/s1/evaluate.py`)
- Experiment Runner → Training Orchestrator: launches runs (evidence: `README.md`)
- Experiment Runner → Generative Baseline: launches baseline (evidence: `README.md`)
- Generative Baseline → Model Resolution: resolves model (evidence: `research/s1/llm_baseline.py`)
- Generative Baseline → Temperature Fitting: fits temperatures (evidence: `research/s1/llm_baseline.py`)
- Generative Baseline → Prediction Evaluation: reuses evaluation (evidence: `research/s1/llm_baseline.py`)
- Data Audit → Paper Analysis: supplies audit data (evidence: `research/paper/scripts/analysis_common.py`)
- Paper Analysis → Temperature Lookup: applies temperatures (evidence: `research/paper/scripts/analysis_common.py`)
- Confidence Cascade → Paper Analysis: reads predictions (evidence: `research/paper/scripts/cascade_v2.py`)
- Results Tables → Run Summaries: summarizes results (evidence: `README.md`)
