Data extraction from text in Prompt Engineering / GenAI - Model Pipeline Trace

Model Pipeline - Data extraction from text

This pipeline takes raw text and finds useful pieces of information inside it. It cleans the text, finds important words or phrases, and then uses a model to pick out the data we want.

Data Flow - 6 Stages

1Raw Text Input

1000 sentences x variable length→Receive raw text data from documents or messages→1000 sentences x variable length

"John bought 3 apples on Monday."

↓

2Text Cleaning

1000 sentences x variable length→Remove punctuation, lowercase all words, remove extra spaces→1000 sentences x variable length

"john bought 3 apples on monday"

↓

3Tokenization

1000 sentences x variable length→Split sentences into words or tokens→1000 sentences x average 7 tokens

["john", "bought", "3", "apples", "on", "monday"]

↓

4Feature Extraction

1000 sentences x average 7 tokens→Convert tokens into numbers using word embeddings→1000 sentences x 7 tokens x 50 features

[[0.12, -0.05, ..., 0.33], ..., [0.01, 0.22, ..., -0.11]]

↓

5Model Prediction

1000 sentences x 7 tokens x 50 features→Use trained model to identify data entities in text→1000 sentences x 7 tokens x 3 classes (entity tags)

["B-PER", "O", "B-QUANTITY", "B-ITEM", "O", "B-DATE"]

↓

6Data Extraction Output

1000 sentences x 7 tokens x 3 classes→Convert tagged tokens into structured data entries→1000 structured records

{"Person": "John", "Quantity": 3, "Item": "apples", "Date": "Monday"}

Training Trace - Epoch by Epoch

1.2 |**************
0.9 |**********
0.7 |*******
0.5 |****
0.4 |***
    +----------------
     1  2  3  4  5 Epochs

Epoch	Loss ↓	Accuracy ↑	Observation
1	1.2	0.55	Model starts learning, loss high, accuracy low
2	0.9	0.68	Loss decreases, accuracy improves
3	0.7	0.75	Model learns important patterns
4	0.5	0.82	Good improvement, model converging
5	0.4	0.87	Loss low, accuracy high, training stable

Prediction Trace - 5 Layers

Layer 1: Input Text

Layer 2: Tokenization

Layer 3: Embedding Layer

Layer 4: Model Prediction

Layer 5: Data Structuring

Model Quiz - 3 Questions

What happens to the text during the 'Text Cleaning' stage?

A Punctuation is removed and text is lowercased

B Text is split into tokens

C Model predicts entity tags

D Structured data is created

Key Insight

This visualization shows how raw text is cleaned, turned into numbers, and then a model learns to find useful data inside. As training goes on, the model gets better at tagging words correctly, which helps us extract structured information from messy text.