Update app.py
Browse files
app.py
CHANGED
|
@@ -1,130 +1,154 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
import gradio as gr
|
| 2 |
import pandas as pd
|
| 3 |
import requests
|
| 4 |
-
import
|
| 5 |
-
|
| 6 |
-
|
| 7 |
-
|
| 8 |
-
|
| 9 |
-
|
| 10 |
-
|
| 11 |
-
|
| 12 |
-
|
| 13 |
-
|
| 14 |
-
|
| 15 |
-
You are an HR compliance
|
| 16 |
-
|
| 17 |
-
|
| 18 |
-
|
| 19 |
-
|
| 20 |
-
-
|
| 21 |
-
-
|
| 22 |
-
-
|
| 23 |
-
-
|
| 24 |
-
-
|
| 25 |
-
|
| 26 |
-
|
| 27 |
-
|
| 28 |
-
|
| 29 |
-
|
| 30 |
-
|
| 31 |
-
|
| 32 |
-
|
| 33 |
-
|
| 34 |
-
|
| 35 |
-
|
| 36 |
-
|
| 37 |
-
|
| 38 |
-
|
| 39 |
-
|
| 40 |
-
|
| 41 |
-
|
| 42 |
-
|
| 43 |
-
|
| 44 |
-
|
| 45 |
-
|
| 46 |
-
|
| 47 |
-
|
| 48 |
-
|
| 49 |
-
|
| 50 |
-
|
| 51 |
-
|
| 52 |
-
|
| 53 |
-
|
| 54 |
-
|
| 55 |
-
|
| 56 |
-
|
| 57 |
-
|
| 58 |
-
|
| 59 |
-
|
| 60 |
-
|
| 61 |
-
|
| 62 |
-
|
| 63 |
-
|
| 64 |
-
|
| 65 |
-
|
| 66 |
-
|
| 67 |
-
|
| 68 |
-
|
| 69 |
-
|
| 70 |
-
|
| 71 |
-
|
| 72 |
-
|
| 73 |
-
|
| 74 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 75 |
|
| 76 |
|
| 77 |
def analyze_file(file):
|
| 78 |
if file is None:
|
| 79 |
return None, None
|
| 80 |
|
| 81 |
-
|
| 82 |
-
df = pd.read_excel(file.name)
|
| 83 |
-
except Exception as e:
|
| 84 |
-
raise gr.Error(f"Failed to read Excel file: {str(e)}")
|
| 85 |
-
|
| 86 |
-
col_names = df.columns.tolist()
|
| 87 |
|
| 88 |
-
#
|
| 89 |
text_col = None
|
| 90 |
-
for
|
| 91 |
-
if
|
| 92 |
-
text_col =
|
| 93 |
break
|
| 94 |
-
|
| 95 |
if text_col is None:
|
| 96 |
-
|
| 97 |
|
| 98 |
-
|
| 99 |
-
|
| 100 |
-
|
| 101 |
-
|
|
|
|
|
|
|
| 102 |
|
| 103 |
-
|
| 104 |
-
|
| 105 |
-
categories_list.append(cat)
|
| 106 |
-
keywords_list.append(key)
|
| 107 |
-
explanations_list.append(exp)
|
| 108 |
|
| 109 |
-
|
| 110 |
-
|
| 111 |
-
df["Explanation"] = explanations_list
|
| 112 |
|
| 113 |
-
|
| 114 |
-
|
| 115 |
|
| 116 |
-
|
|
|
|
| 117 |
|
| 118 |
-
except Exception as e:
|
| 119 |
-
raise gr.Error(f"Columns: {col_names}. Error: {str(e)}")
|
| 120 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 121 |
|
| 122 |
demo = gr.Interface(
|
| 123 |
fn=analyze_file,
|
| 124 |
-
inputs=gr.File(label="Upload Job Posts Excel"),
|
| 125 |
-
outputs=[
|
| 126 |
-
|
| 127 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 128 |
)
|
| 129 |
|
| 130 |
-
demo.launch(server_name="0.0.0.0", server_port=7860)
|
|
|
|
| 1 |
+
import os
|
| 2 |
+
import re
|
| 3 |
+
import json
|
| 4 |
+
import time
|
| 5 |
import gradio as gr
|
| 6 |
import pandas as pd
|
| 7 |
import requests
|
| 8 |
+
from discrimination_model import detect_discrimination
|
| 9 |
+
|
| 10 |
+
# ββ LLM backend selector βββββββββββββββββββββββββββββββββββββββββββββββββββββ
|
| 11 |
+
# Set ONE of these in Space Settings β Variables and secrets:
|
| 12 |
+
# ANTHROPIC_API_KEY β uses Claude (Option A, higher accuracy)
|
| 13 |
+
# HF_TOKEN β uses Mistral via HF Inference API (Option B, free)
|
| 14 |
+
# If neither is set, falls back to the original keyword-based rules.
|
| 15 |
+
|
| 16 |
+
ANTHROPIC_KEY = os.environ.get("ANTHROPIC_API_KEY", "")
|
| 17 |
+
HF_TOKEN = os.environ.get("HF_TOKEN", "")
|
| 18 |
+
|
| 19 |
+
SYSTEM_PROMPT = """You are an expert HR compliance analyst.
|
| 20 |
+
Analyse the job posting and identify discriminatory elements from this list:
|
| 21 |
+
- Gender: gendered job titles or language implying a preferred gender
|
| 22 |
+
- Age: explicit age limits or age-suggestive language
|
| 23 |
+
- Origin: nationality, ethnicity, or geographic residency requirements
|
| 24 |
+
- Mobility: unjustified driving licence or own vehicle requirement
|
| 25 |
+
- Salary: undisclosed or vague compensation (e.g. "to be defined", "commensurate with experience")
|
| 26 |
+
- Standing: appearance, weight, height, or "professional look" requirements
|
| 27 |
+
- PPE: candidate required to supply own safety equipment
|
| 28 |
+
- Role clarity: vague/contradictory description masking discriminatory intent
|
| 29 |
+
- None: if fully compliant
|
| 30 |
+
|
| 31 |
+
Return ONLY valid JSON, no markdown, no extra text:
|
| 32 |
+
{"discriminatory_elements": ["Element1"], "explanation": "brief reason per element"}"""
|
| 33 |
+
|
| 34 |
+
|
| 35 |
+
def llm_detect(text: str) -> tuple[str, str]:
|
| 36 |
+
"""Call LLM backend. Returns (categories_str, explanation_str)."""
|
| 37 |
+
|
| 38 |
+
# ββ Option A: Claude βββββββββββββββββββββββββββββββββββββββββββββββββββββ
|
| 39 |
+
if ANTHROPIC_KEY:
|
| 40 |
+
try:
|
| 41 |
+
resp = requests.post(
|
| 42 |
+
"https://api.anthropic.com/v1/messages",
|
| 43 |
+
headers={
|
| 44 |
+
"x-api-key": ANTHROPIC_KEY,
|
| 45 |
+
"anthropic-version": "2023-06-01",
|
| 46 |
+
"content-type": "application/json",
|
| 47 |
+
},
|
| 48 |
+
json={
|
| 49 |
+
"model": "claude-haiku-4-5-20251001", # fast + cheap
|
| 50 |
+
"max_tokens": 512,
|
| 51 |
+
"system": SYSTEM_PROMPT,
|
| 52 |
+
"messages": [{"role": "user", "content": f"Job posting:\n\n{text}"}],
|
| 53 |
+
},
|
| 54 |
+
timeout=30,
|
| 55 |
+
)
|
| 56 |
+
resp.raise_for_status()
|
| 57 |
+
raw = resp.json()["content"][0]["text"].strip()
|
| 58 |
+
data = json.loads(raw)
|
| 59 |
+
elements = data.get("discriminatory_elements", ["None"])
|
| 60 |
+
explanation = data.get("explanation", "")
|
| 61 |
+
return ", ".join(elements), explanation
|
| 62 |
+
except Exception as e:
|
| 63 |
+
return "Error (Claude)", str(e)
|
| 64 |
+
|
| 65 |
+
# ββ Option B: HF Inference API (Mistral-7B) ββββββββββββββββββββββββββββββ
|
| 66 |
+
if HF_TOKEN:
|
| 67 |
+
try:
|
| 68 |
+
prompt = (
|
| 69 |
+
f"<s>[INST] {SYSTEM_PROMPT}\n\nJob posting:\n\n{text} [/INST]"
|
| 70 |
+
)
|
| 71 |
+
resp = requests.post(
|
| 72 |
+
"https://api-inference.huggingface.co/models/mistralai/Mistral-7B-Instruct-v0.3",
|
| 73 |
+
headers={"Authorization": f"Bearer {HF_TOKEN}"},
|
| 74 |
+
json={"inputs": prompt, "parameters": {"max_new_tokens": 300, "temperature": 0.1}},
|
| 75 |
+
timeout=60,
|
| 76 |
+
)
|
| 77 |
+
resp.raise_for_status()
|
| 78 |
+
output = resp.json()
|
| 79 |
+
generated = output[0]["generated_text"] if isinstance(output, list) else str(output)
|
| 80 |
+
# Extract JSON from the response
|
| 81 |
+
match = re.search(r'\{.*\}', generated, re.DOTALL)
|
| 82 |
+
if match:
|
| 83 |
+
data = json.loads(match.group())
|
| 84 |
+
elements = data.get("discriminatory_elements", ["None"])
|
| 85 |
+
explanation = data.get("explanation", "")
|
| 86 |
+
return ", ".join(elements), explanation
|
| 87 |
+
return "Parsing error", generated[:200]
|
| 88 |
+
except Exception as e:
|
| 89 |
+
return "Error (HF)", str(e)
|
| 90 |
+
|
| 91 |
+
# ββ Fallback: keyword rules βββββββββββββββββββββββββββββββββββββββββββββββ
|
| 92 |
+
return detect_discrimination(text)
|
| 93 |
|
| 94 |
|
| 95 |
def analyze_file(file):
|
| 96 |
if file is None:
|
| 97 |
return None, None
|
| 98 |
|
| 99 |
+
df = pd.read_excel(file)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 100 |
|
| 101 |
+
# Accept either column name convention
|
| 102 |
text_col = None
|
| 103 |
+
for candidate in ["Job post text", "job_post", "text", "job post"]:
|
| 104 |
+
if candidate in df.columns:
|
| 105 |
+
text_col = candidate
|
| 106 |
break
|
|
|
|
| 107 |
if text_col is None:
|
| 108 |
+
return gr.update(value="β Could not find job text column. Expected: 'Job post text' or 'job_post'"), None
|
| 109 |
|
| 110 |
+
categories, explanations = [], []
|
| 111 |
+
for text in df[text_col]:
|
| 112 |
+
cat, exp = llm_detect(str(text))
|
| 113 |
+
categories.append(cat)
|
| 114 |
+
explanations.append(exp)
|
| 115 |
+
time.sleep(0.3) # gentle rate-limit buffer
|
| 116 |
|
| 117 |
+
df["Discrimination_Type"] = categories
|
| 118 |
+
df["Explanation"] = explanations
|
|
|
|
|
|
|
|
|
|
| 119 |
|
| 120 |
+
output_path = "/tmp/results.xlsx"
|
| 121 |
+
df.to_excel(output_path, index=False)
|
|
|
|
| 122 |
|
| 123 |
+
backend = "Claude API" if ANTHROPIC_KEY else ("HF Inference API" if HF_TOKEN else "Keyword rules")
|
| 124 |
+
print(f"Analysis complete β backend: {backend}, rows: {len(df)}")
|
| 125 |
|
| 126 |
+
return df[["Job post ID" if "Job post ID" in df.columns else df.columns[0],
|
| 127 |
+
text_col, "Discrimination_Type", "Explanation"]], output_path
|
| 128 |
|
|
|
|
|
|
|
| 129 |
|
| 130 |
+
# ββ Gradio UI βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
|
| 131 |
+
backend_label = (
|
| 132 |
+
"π€ Claude API" if ANTHROPIC_KEY
|
| 133 |
+
else ("π€ HF Inference API (Mistral-7B)" if HF_TOKEN
|
| 134 |
+
else "π Keyword rules (no API key set)")
|
| 135 |
+
)
|
| 136 |
|
| 137 |
demo = gr.Interface(
|
| 138 |
fn=analyze_file,
|
| 139 |
+
inputs=gr.File(label="Upload Job Posts Excel (.xlsx)"),
|
| 140 |
+
outputs=[
|
| 141 |
+
gr.Dataframe(label=f"Results β backend: {backend_label}"),
|
| 142 |
+
gr.File(label="Download results.xlsx"),
|
| 143 |
+
],
|
| 144 |
+
title="π AI Job Post Discrimination Checker",
|
| 145 |
+
description=(
|
| 146 |
+
"Upload the job posts Excel file. "
|
| 147 |
+
"The tool detects: **Gender Β· Age Β· Origin Β· Mobility Β· Salary Β· "
|
| 148 |
+
"Standing Β· PPE Β· Role clarity**.\n\n"
|
| 149 |
+
f"Active backend: `{backend_label}`"
|
| 150 |
+
),
|
| 151 |
+
allow_flagging="never",
|
| 152 |
)
|
| 153 |
|
| 154 |
+
demo.launch(server_name="0.0.0.0", server_port=7860)
|