Person 1

• Collect a small dataset (FAERS CSV + PubMed abstracts).
• Write script to clean text (lowercase, remove symbols, sentence split with SpaCy).
• Store cleaned text in data/cleaned/.

---

Person 2

• Implement NER (SciSpacy/Med7) to detect drugs and medical events.
• Output results as JSON: {drug: "X", event: "Y"}.
• Save outputs in results/extractions.json.

---

Person 3

• Build a simple sentence classifier (ADE vs Non-ADE).
• Start with TF-IDF + logistic regression (scikit-learn).
• Expose it as a Python function: classify_sentence(text) -> label, confidence.

---

Person 4

• Set up FastAPI backend.

• Create endpoints:

  • /extract → runs Person 2’s pipeline.
  • /classify → runs Person 3’s classifier.

• Make sure it returns JSON responses.

---

Person 5

• Build a simple React/Next.js frontend.
• Text box → send text to backend /extract + /classify.
• Highlight drugs in blue, ADEs in red.

---

Person 6

• Take Person 2’s extraction results.

• Compute basic stats:

  • Frequency of ADEs.
  • Top drugs with ADEs.

• Show results as bar chart / table in frontend.

---

Sequence

1. Person 1 finishes data prep.
2. Person 2 + Person 3 use cleaned data to build models.
3. Person 4 integrates extraction + classification into backend.
4. Person 5 connects frontend to backend.
5. Person 6 adds stats/visualization once extraction results are available.

---

Final Flow

• Data → NER → Classification → API → UI → Stats.