LCCC Condition Extraction

Extract LCCC (liquid chromatography at critical conditions) experiment conditions from raw paper text with minimal hardcoding.

Quick Start

python runner.py rawData/paper.txt --no-validation

Output is written to extracted_lccc_data.json and contains LCCC_conditions.

Example output shape:

{
  "metadata": { "..." : "..." },
  "LCCC_conditions": [
    {
      "critical_polymer_unit": "EO",
      "SP_FIELDS": { "column_name": "..." },
      "SOL_FIELDS": { "solvent": "...", "ratio": "...", "ratio_units": "..." },
      "TECH_FIELDS": { "temperature": "...", "flow_rate": "..." },
      "separation_behavior": { "critical_block_behavior": "...", "purpose": "..." }
    }
  ]
}

Output Mode

To keep the JSON output clean (no evidence/debug fields), use the default compact mode. You can switch to full to keep all evidence/debug fields for troubleshooting.

# Clean output (default)
python runner.py rawData/paper.txt --output-mode compact

# Full output with evidence/debug fields
python runner.py rawData/paper.txt --output-mode full

LLM Extraction (Ollama)

The pipeline uses local Ollama models as the primary extraction engine. All extracted values are validated against the source text.

# Single model
python runner.py rawData/paper.txt --llm-provider ollama --llm-model qwen2.5:7b

# Multi-model with consensus
python runner.py rawData/paper.txt --llm-models "qwen3:8b,kimi-k2:7b" --llm-consensus 2

LLM configuration via env vars (optional):

• LCCC_LLM_PROVIDER (default: ollama)
• LCCC_LLM_MODEL (default: qwen2.5:7b)
• LCCC_LLM_MODELS (comma list, e.g. qwen3:8b,kimi-k2:7b)
• LCCC_LLM_CONSENSUS (min model agreement)
• LCCC_LLM_HOST (default: http://127.0.0.1:11434)
• LCCC_LLM_TIMEOUT (seconds)
• LCCC_LLM_MAX_CONTEXT (characters)
• LCCC_LLM_DEBUG (set to 1 for per-chunk logging)

Project Layout

• runner.py: CLI runner and human-readable summary printer.
• pipeline.py: Compatibility wrapper (keeps the old from pipeline import ... import path working).
• lccc_extractor/pipeline.py: Pipeline orchestration (run_pipeline, save_json).
• lccc_extractor/llm/: LLM client + extractor.
• lccc_extractor/polymers/: Polymer catalog + validation.
• lccc_extractor/text/: Text normalization + paragraph splitting utilities.
• lccc_extractor/linking.py: Condition linking + validation.
• polymerSubject.py: Compatibility wrapper for legacy imports.

Environment Variables

• LCCC_LOG_LEVEL: DEBUG, INFO (default), WARNING, ...