feat: Initialize AutoForm AI project structure
Sets up the basic project structure for AutoForm AI, including: - Vite for build tooling and development server. - React and ReactDOM for the UI. - TypeScript for static typing. - Essential dependencies for PDF manipulation (jspdf, pdf-lib) and AI integration (@google/genai). - Basic HTML structure and styling. - Component definitions and service interfaces for future development. - A README with local development instructions.
This commit is contained in:
Kenearos
parent
f1f796c9ca
commit
d2ea8a0cd4
14 changed files with 1217 additions and 8 deletions
161
services/geminiService.ts
Normal file
161
services/geminiService.ts
Normal file
|
|
@ -0,0 +1,161 @@
|
|||
import { GoogleGenAI, Type, Schema } from "@google/genai";
|
||||
import { FileData, FormResponse } from "../types";
|
||||
import { PdfFieldInfo } from "./pdfService";
|
||||
|
||||
const ai = new GoogleGenAI({ apiKey: process.env.API_KEY });
|
||||
|
||||
const responseSchema: Schema = {
|
||||
type: Type.OBJECT,
|
||||
properties: {
|
||||
summary: {
|
||||
type: Type.STRING,
|
||||
description: "A brief summary of what document was processed."
|
||||
},
|
||||
fields: {
|
||||
type: Type.ARRAY,
|
||||
items: {
|
||||
type: Type.OBJECT,
|
||||
properties: {
|
||||
key: {
|
||||
type: Type.STRING,
|
||||
description: "The PDF field name (if available)."
|
||||
},
|
||||
label: {
|
||||
type: Type.STRING,
|
||||
description: "A human-readable label for the field."
|
||||
},
|
||||
value: {
|
||||
type: Type.STRING,
|
||||
description: "The value to fill. For checkboxes, use 'true'/'false' or 'X'."
|
||||
},
|
||||
sourceContext: {
|
||||
type: Type.STRING,
|
||||
description: "The exact snippet of text from the source document used to derive this value. Used for user verification."
|
||||
},
|
||||
coordinates: {
|
||||
type: Type.OBJECT,
|
||||
description: "REQUIRED if no specific PDF field names are provided. Visual location to draw text.",
|
||||
properties: {
|
||||
pageIndex: { type: Type.INTEGER, description: "0-based page index" },
|
||||
x: { type: Type.INTEGER, description: "Horizontal position (0-1000) from Left" },
|
||||
y: { type: Type.INTEGER, description: "Vertical position (0-1000) from Top" }
|
||||
},
|
||||
required: ["pageIndex", "x", "y"]
|
||||
},
|
||||
validation: {
|
||||
type: Type.OBJECT,
|
||||
properties: {
|
||||
status: {
|
||||
type: Type.STRING,
|
||||
description: "VALID, WARNING, or INVALID."
|
||||
},
|
||||
message: {
|
||||
type: Type.STRING,
|
||||
description: "Validation message explaining any issues or uncertainty."
|
||||
},
|
||||
suggestion: {
|
||||
type: Type.STRING,
|
||||
description: "Alternative value suggestion if the extracted value is uncertain."
|
||||
}
|
||||
},
|
||||
required: ["status"]
|
||||
}
|
||||
},
|
||||
required: ["label", "value", "validation"]
|
||||
}
|
||||
}
|
||||
},
|
||||
required: ["fields", "summary"]
|
||||
};
|
||||
|
||||
export const processDocuments = async (
|
||||
blankForm: FileData,
|
||||
sourceDocument: FileData,
|
||||
pdfFields: PdfFieldInfo[] = []
|
||||
): Promise<FormResponse> => {
|
||||
|
||||
const formPart = {
|
||||
inlineData: {
|
||||
data: blankForm.base64,
|
||||
mimeType: blankForm.type,
|
||||
},
|
||||
};
|
||||
|
||||
const sourcePart = {
|
||||
inlineData: {
|
||||
data: sourceDocument.base64,
|
||||
mimeType: sourceDocument.type,
|
||||
},
|
||||
};
|
||||
|
||||
let systemPrompt = `
|
||||
ROLE: Intelligent Document Processing AI (Verification Expert).
|
||||
TASK: Extract data from the SOURCE DOCUMENT and map it to the BLANK TARGET FORM.
|
||||
|
||||
CRITICAL INSTRUCTION: You must verify every extraction. If a value is ambiguous, plausibility is low, or you are guessing, set validation.status to 'WARNING' and explain why in validation.message.
|
||||
`;
|
||||
|
||||
if (pdfFields.length > 0) {
|
||||
const fieldList = pdfFields.map(f => `"${f.name}" (${f.type})`).join(", ");
|
||||
systemPrompt += `
|
||||
MODE: FILLABLE PDF (AcroForm).
|
||||
The target form has specific embedded fields.
|
||||
Map extracted data to these exact field IDs: [${fieldList}].
|
||||
Return the 'key' property matching the field ID.
|
||||
`;
|
||||
} else {
|
||||
systemPrompt += `
|
||||
MODE: VISUAL FILLING (Flat/XFA/Scan).
|
||||
The target form DOES NOT have accessible digital fields.
|
||||
You must VISUALLY locate where the text should be written.
|
||||
|
||||
For every field you identify on the TARGET FORM:
|
||||
1. Extract the corresponding value from the SOURCE DOCUMENT.
|
||||
2. Estimate the VISUAL COORDINATES [pageIndex, x, y] where the text should start.
|
||||
- 'x' and 'y' are on a scale of 0 to 1000.
|
||||
- (0,0) is the top-left corner of the page.
|
||||
- (1000,1000) is the bottom-right corner.
|
||||
- Align text slightly above lines or inside boxes.
|
||||
|
||||
For checkboxes: If true/yes, the value should be "X" placed inside the box.
|
||||
`;
|
||||
}
|
||||
|
||||
systemPrompt += `
|
||||
VALIDATION RULES:
|
||||
1. Dates: Ensure format matches the form (e.g. DD.MM.YYYY).
|
||||
2. Checkboxes: Only mark if explicitly supported by source.
|
||||
3. Missing Data: If a field is not found in source, leave 'value' empty and set status 'VALID'. Do not hallucinate.
|
||||
4. Source Context: Always populate 'sourceContext' with the exact text snippet from the source document that justifies your extraction.
|
||||
`;
|
||||
|
||||
try {
|
||||
const modelId = "gemini-3-flash-preview";
|
||||
|
||||
const response = await ai.models.generateContent({
|
||||
model: modelId,
|
||||
contents: {
|
||||
parts: [
|
||||
formPart,
|
||||
{ text: "This is the BLANK TARGET FORM." },
|
||||
sourcePart,
|
||||
{ text: "This is the SOURCE DOCUMENT." },
|
||||
]
|
||||
},
|
||||
config: {
|
||||
responseMimeType: "application/json",
|
||||
responseSchema: responseSchema,
|
||||
systemInstruction: systemPrompt
|
||||
}
|
||||
});
|
||||
|
||||
const text = response.text;
|
||||
if (!text) throw new Error("No response from Gemini");
|
||||
|
||||
return JSON.parse(text) as FormResponse;
|
||||
|
||||
} catch (error) {
|
||||
console.error("Gemini API Error:", error);
|
||||
throw error;
|
||||
}
|
||||
};
|
||||
104
services/pdfService.ts
Normal file
104
services/pdfService.ts
Normal file
|
|
@ -0,0 +1,104 @@
|
|||
import { PDFDocument, PDFTextField, PDFCheckBox, StandardFonts, rgb } from 'pdf-lib';
|
||||
import { ExtractedField } from '../types';
|
||||
|
||||
export interface PdfFieldInfo {
|
||||
name: string;
|
||||
type: string;
|
||||
}
|
||||
|
||||
export const getPdfFields = async (base64: string): Promise<PdfFieldInfo[]> => {
|
||||
try {
|
||||
const pdfDoc = await PDFDocument.load(base64);
|
||||
const form = pdfDoc.getForm();
|
||||
const fields = form.getFields();
|
||||
return fields.map(f => ({
|
||||
name: f.getName(),
|
||||
type: f.constructor.name
|
||||
}));
|
||||
} catch (error) {
|
||||
console.warn("Failed to extract PDF fields", error);
|
||||
return [];
|
||||
}
|
||||
};
|
||||
|
||||
export const createFilledPdf = async (base64: string, fields: ExtractedField[], isFillable: boolean): Promise<Uint8Array> => {
|
||||
const pdfDoc = await PDFDocument.load(base64);
|
||||
const pages = pdfDoc.getPages();
|
||||
const font = await pdfDoc.embedFont(StandardFonts.Helvetica);
|
||||
|
||||
if (isFillable) {
|
||||
try {
|
||||
const form = pdfDoc.getForm();
|
||||
const fieldMap: Record<string, string> = {};
|
||||
fields.forEach(f => {
|
||||
if (f.key) fieldMap[f.key] = f.value;
|
||||
});
|
||||
|
||||
for (const [key, value] of Object.entries(fieldMap)) {
|
||||
try {
|
||||
const field = form.getField(key);
|
||||
if (!field) continue;
|
||||
|
||||
if (field instanceof PDFTextField) {
|
||||
field.setText(String(value));
|
||||
} else if (field instanceof PDFCheckBox) {
|
||||
const isChecked = String(value).toLowerCase() === 'true' || String(value).toLowerCase() === 'yes';
|
||||
if (isChecked) field.check();
|
||||
else field.uncheck();
|
||||
}
|
||||
} catch (e) {
|
||||
// Field might be read-only or tricky
|
||||
}
|
||||
}
|
||||
} catch (e) {
|
||||
console.warn("Error filling form fields", e);
|
||||
}
|
||||
} else {
|
||||
// VISUAL OVERLAY MODE
|
||||
// Iterate through fields and draw them at specific coordinates
|
||||
|
||||
for (const field of fields) {
|
||||
// Skip if no value or no coordinates
|
||||
if (!field.value || !field.coordinates) continue;
|
||||
|
||||
const { pageIndex, x, y } = field.coordinates;
|
||||
|
||||
// Safety check for page index
|
||||
if (pageIndex < 0 || pageIndex >= pages.length) continue;
|
||||
|
||||
const page = pages[pageIndex];
|
||||
const { width, height } = page.getSize();
|
||||
|
||||
// Convert 0-1000 coordinates to PDF Point coordinates
|
||||
// PDF (0,0) is bottom-left.
|
||||
// API (0,0) is top-left.
|
||||
// x = (x / 1000) * width
|
||||
// y = height - (y / 1000) * height
|
||||
|
||||
const pdfX = (x / 1000) * width;
|
||||
const pdfY = height - (y / 1000) * height;
|
||||
|
||||
// Adjust slightly for font height (text is drawn from baseline)
|
||||
// A small nudge down (subtract from Y) helps align with lines usually.
|
||||
const adjustedY = pdfY - 4;
|
||||
|
||||
try {
|
||||
page.drawText(field.value, {
|
||||
x: pdfX,
|
||||
y: adjustedY,
|
||||
size: 10,
|
||||
font: font,
|
||||
color: rgb(0, 0, 0),
|
||||
});
|
||||
} catch (e) {
|
||||
console.warn(`Failed to draw field ${field.label}`, e);
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
return await pdfDoc.save();
|
||||
};
|
||||
|
||||
export const fillPdf = async (base64: string, fieldValues: Record<string, string | boolean>): Promise<Uint8Array> => {
|
||||
return new Uint8Array();
|
||||
};
|
||||
Loading…
Add table
Add a link
Reference in a new issue