File size: 20,373 Bytes
6ba0bd9
 
 
92fdc45
0530a8b
e776a35
 
0530a8b
03fc26b
0530a8b
 
e776a35
0530a8b
 
 
 
6ba0bd9
 
 
92fdc45
6ba0bd9
0530a8b
 
 
 
92fdc45
0530a8b
6ba0bd9
 
0530a8b
6ba0bd9
 
 
e7d38dd
92fdc45
6ba0bd9
92fdc45
0530a8b
03fc26b
 
0530a8b
 
 
 
 
 
 
 
 
e776a35
0530a8b
 
 
 
 
 
 
 
e776a35
 
0530a8b
 
 
 
 
 
 
 
 
e776a35
 
0530a8b
3c2c0f5
0530a8b
 
 
 
 
 
e776a35
 
0530a8b
e776a35
0530a8b
 
 
 
 
 
e776a35
 
0530a8b
 
 
 
 
 
 
 
e776a35
 
0530a8b
 
 
 
 
 
 
 
e776a35
 
0530a8b
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
e776a35
 
 
0530a8b
 
 
 
 
 
 
 
 
 
 
 
e776a35
0530a8b
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
92fdc45
 
 
 
 
 
 
 
 
0530a8b
92fdc45
 
 
 
0530a8b
 
e776a35
0530a8b
 
e776a35
0530a8b
 
 
e776a35
 
0530a8b
e776a35
0530a8b
 
 
e776a35
 
0530a8b
e776a35
0530a8b
 
e776a35
0530a8b
e776a35
 
 
 
 
3c2c0f5
0530a8b
e776a35
 
 
 
 
 
 
 
3c2c0f5
e776a35
 
 
0530a8b
3c2c0f5
e776a35
3c2c0f5
 
 
0530a8b
 
3c2c0f5
0530a8b
e776a35
0530a8b
e776a35
 
3c2c0f5
 
 
 
 
0530a8b
 
 
 
 
 
3c2c0f5
e776a35
0530a8b
 
 
e776a35
0530a8b
e776a35
0530a8b
 
e776a35
 
0530a8b
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
03fc26b
0530a8b
e776a35
 
0530a8b
 
 
 
 
6ba0bd9
0530a8b
e776a35
0530a8b
 
e776a35
92fdc45
0530a8b
 
 
 
 
 
92fdc45
 
0530a8b
 
 
 
 
 
 
 
92fdc45
0530a8b
 
 
 
 
 
 
92fdc45
0530a8b
 
e776a35
0530a8b
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
92fdc45
6ba0bd9
0530a8b
92fdc45
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
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
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
#!/usr/bin/env python3
"""
IRS Publication 1075 Compliance Assistant (Gradio)

Features
- Upload a PDF or DOCX policy/security document and run heuristic checks aligned to IRS Pub. 1075 themes.
- Generate a structured compliance report with findings, gaps, and actionable recommendations.
- Ask detailed Pub. 1075 questions and get answers grounded in the OFFICIAL PDF only:
  https://www.irs.gov/pub/irs-pdf/p1075.pdf
  The app downloads the PDF at runtime (if internet is available), builds a page-level index,
  and cites specific page numbers and the most relevant passages.

Notes
- Do NOT upload real FTI; use redacted/sample docs. This is guidance only, not legal advice.
- Files are handled in memory; the generated report file is written to /tmp for download.
- If the Space has no internet access, Q&A will fall back to a minimal local summary and warn the user.
"""

import io
import os
import re
import json
import time
import math
import tempfile
from datetime import datetime
from typing import List, Dict, Any, Tuple, Optional

import gradio as gr
import requests
from sklearn.feature_extraction.text import TfidfVectorizer
from sklearn.metrics.pairwise import cosine_similarity

from PyPDF2 import PdfReader
from docx import Document as DocxDocument

APP_TITLE = "IRS Pub. 1075 Compliance Assistant"
APP_TAGLINE = "Upload a policy/security document for heuristic checks and ask detailed Pub. 1075 questions with page citations."
IRS_PUB_1075_URL = "https://www.irs.gov/pub/irs-pdf/p1075.pdf"

WARNING_BANNER = (
    "⚠️ Do NOT upload real FTI. This tool references only IRS Publication 1075 "
    f"({IRS_PUB_1075_URL}). It provides guidance only, not legal/compliance advice."
)

# --------------------------------------------------------------------------------------
# Knowledge themes for heuristic checks (no external sources beyond Pub. 1075)
# --------------------------------------------------------------------------------------
CHECKS = [
    {
        "id": "chk_encryption",
        "title": "Cryptographic Protections (At Rest & In Transit)",
        "patterns_any": [
            r"\bencrypt(ed|ion|ing)\b", r"\bTLS\s*1\.2\b", r"\bTLS\s*1\.3\b",
            r"\bcryptograph(y|ic)\b", r"\bkey management\b", r"\bkey rotation\b", r"\bFIPS\b"
        ],
        "recommendation": "Document approved cryptography (preferring FIPS-validated modules if applicable), TLS 1.2+ for transmission, and key management (generation, storage, rotation).",
        "section": "Encryption & Cryptographic Protections"
    },
    {
        "id": "chk_access",
        "title": "Access Control, Least Privilege & MFA",
        "patterns_any": [
            r"\bMFA\b", r"\bmulti-?factor\b", r"\bleast privilege\b", r"\brole-?based\b",
            r"\baccess control\b", r"\bprivileged\b", r"\badmin(istrative)? access\b",
            r"\baccount (lifecycle|provisioning|deprovisioning|review)\b"
        ],
        "recommendation": "Enforce least privilege and role-based access. Require MFA for remote and admin access. Define account provisioning, periodic reviews, and timely deprovisioning.",
        "section": "Access Control & Multi-Factor Authentication"
    },
    {
        "id": "chk_incident",
        "title": "Incident Response & Reporting",
        "patterns_any": [
            r"\bincident response\b", r"\bbreach\b", r"\bcontainment\b", r"\beradication\b",
            r"\brecovery\b", r"\bpost-incident\b", r"\bnotification\b", r"\breport(ing)?\b"
        ],
        "recommendation": "Define and test procedures across detection, escalation, containment, eradication, recovery, and post-incident review. Include notification/reporting timelines.",
        "section": "Incident Response"
    },
    {
        "id": "chk_audit",
        "title": "Audit Logging & Monitoring",
        "patterns_any": [
            r"\baudit(ing)?\b", r"\blog(s|ging)?\b", r"\bretention\b",
            r"\badmin(istrative)? actions?\b", r"\bintegrity\b", r"\btamper\b", r"\bSIEM\b"
        ],
        "recommendation": "Record access and administrative actions; protect logs from tampering; define retention and review procedures.",
        "section": "Audit & Accountability"
    },
    {
        "id": "chk_media",
        "title": "Media Protection & Sanitization",
        "patterns_any": [
            r"\bmedia (sanitization|protection|handling|labeling)\b",
            r"\bdispose|disposal|destruct(ion)?\b", r"\bchain of custody\b", r"\btransport\b"
        ],
        "recommendation": "Define labeling, handling, transport, and sanitization/disposal procedures for media that may contain FTI.",
        "section": "Media Protection"
    },
    {
        "id": "chk_config",
        "title": "Configuration & Vulnerability Management",
        "patterns_any": [
            r"\bconfiguration management\b", r"\bbaseline\b", r"\bchange control\b",
            r"\bpatch(ing)?\b", r"\bvulnerabilit(y|ies)\b", r"\bscan(ning)?\b", r"\bremediation\b"
        ],
        "recommendation": "Maintain baselines, change control, and patch/vulnerability processes. Track remediation timelines.",
        "section": "Configuration & Vulnerability Management"
    },
    {
        "id": "chk_contingency",
        "title": "Contingency Planning & Backup",
        "patterns_any": [
            r"\bcontingency\b", r"\bdisaster recovery\b", r"\bDRP\b", r"\bBCP\b",
            r"\bbackup(s)?\b", r"\brestore\b", r"\btabletop\b", r"\bexercise\b"
        ],
        "recommendation": "Develop, maintain, and test contingency/DR plans and secure, tested backups with documented RTO/RPO.",
        "section": "Contingency Planning"
    },
    {
        "id": "chk_physical",
        "title": "Physical & Environmental Security",
        "patterns_any": [
            r"\bphysical security\b", r"\bdata center\b", r"\bvisitor\b", r"\bbadge\b",
            r"\bperimeter\b", r"\bcamera\b", r"\benvironmental\b"
        ],
        "recommendation": "Restrict physical access; implement visitor controls and appropriate environmental safeguards.",
        "section": "Physical & Environmental Security"
    },
    {
        "id": "chk_training",
        "title": "Security Awareness & Training",
        "patterns_any": [
            r"\bsecurity awareness\b", r"\btraining\b", r"\bannual\b", r"\brole-based\b",
            r"\bprivacy training\b", r"\bFTI training\b"
        ],
        "recommendation": "Provide initial and periodic training; include role-specific content for admins and developers.",
        "section": "Awareness & Training"
    },
]

# --------------------------------------------------------------------------------------
# PDF fetching and page-level indexing for detailed, section-specific Q&A
# --------------------------------------------------------------------------------------
PDF_CACHE_PATH = os.path.join(tempfile.gettempdir(), "irs_pub1075.pdf")
INDEX_JSON_PATH = os.path.join(tempfile.gettempdir(), "irs_pub1075_index.json")

PAGE_TEXTS: List[str] = []
PAGE_HEADINGS: List[str] = []
PAGE_VECTORIZER: Optional[TfidfVectorizer] = None
PAGE_TFIDF = None
PDF_AVAILABLE: bool = False
PDF_PAGES: int = 0

def _first_nonempty_line(text: str) -> str:
    for line in (text or "").splitlines():
        ln = line.strip()
        if ln:
            return ln[:120]
    return "Untitled section"

def _split_sentences(text: str) -> List[str]:
    # Simple sentence splitter
    parts = re.split(r'(?<=[\.\?!])\s+(?=[A-Z0-9])', text.strip())
    # Filter and trim
    return [p.strip() for p in parts if p.strip()]

def _download_pdf_if_needed() -> bool:
    # Return True if available (downloaded or already cached)
    try:
        if os.path.exists(PDF_CACHE_PATH) and os.path.getsize(PDF_CACHE_PATH) > 0:
            return True
        resp = requests.get(IRS_PUB_1075_URL, timeout=30)
        resp.raise_for_status()
        with open(PDF_CACHE_PATH, "wb") as f:
            f.write(resp.content)
        return True
    except Exception:
        return False

def _build_page_index() -> bool:
    global PAGE_TEXTS, PAGE_HEADINGS, PAGE_VECTORIZER, PAGE_TFIDF, PDF_PAGES
    try:
        with open(PDF_CACHE_PATH, "rb") as f:
            reader = PdfReader(f)
            PAGE_TEXTS = []
            PAGE_HEADINGS = []
            PDF_PAGES = len(reader.pages)
            for i in range(PDF_PAGES):
                try:
                    txt = reader.pages[i].extract_text() or ""
                except Exception:
                    txt = ""
                PAGE_TEXTS.append(txt)
                PAGE_HEADINGS.append(_first_nonempty_line(txt))

        # Build TF-IDF over pages (page-level retrieval)
        PAGE_VECTORIZER = TfidfVectorizer(stop_words="english")
        PAGE_TFIDF = PAGE_VECTORIZER.fit_transform(PAGE_TEXTS)
        # Save a tiny index manifest (optional)
        with open(INDEX_JSON_PATH, "w") as jf:
            json.dump({"pages": PDF_PAGES, "cached_at": time.time()}, jf)
        return True
    except Exception:
        PAGE_TEXTS, PAGE_HEADINGS, PAGE_VECTORIZER, PAGE_TFIDF = [], [], None, None
        return False

def ensure_pdf_index_ready() -> bool:
    global PDF_AVAILABLE
    if PAGE_TFIDF is not None and PAGE_VECTORIZER is not None and PAGE_TEXTS:
        PDF_AVAILABLE = True
        return True
    if not _download_pdf_if_needed():
        PDF_AVAILABLE = False
        return False
    ok = _build_page_index()
    PDF_AVAILABLE = ok
    return ok

def search_pub1075_pages(query: str, top_k: int = 5) -> List[Dict[str, Any]]:
    """
    Returns a list of dicts: {page, heading, score, snippets: [ ... ] }
    Each 'snippets' item is a short sentence-level excerpt from that page.
    """
    if not ensure_pdf_index_ready():
        return []

    q_vec = PAGE_VECTORIZER.transform([query])
    sims = cosine_similarity(q_vec, PAGE_TFIDF).flatten()
    order = sims.argsort()[::-1][:max(1, top_k)]

    results = []
    for idx in order:
        page_text = PAGE_TEXTS[idx]
        heading = PAGE_HEADINGS[idx]
        sentences = _split_sentences(page_text)
        # Score sentences by simple TF-IDF dot with the same vectorizer (fallback: substring hit count)
        try:
            sent_vecs = PAGE_VECTORIZER.transform(sentences)
            s_sims = cosine_similarity(q_vec, sent_vecs).flatten()
            top_sent_idx = s_sims.argsort()[::-1][:3]
            best_snips = [sentences[i] for i in top_sent_idx if sentences[i]]
        except Exception:
            # Fallback: choose sentences containing query terms
            q_terms = [t for t in re.findall(r"\w+", query.lower()) if len(t) > 2]
            scored = []
            for s in sentences:
                score = sum(1 for t in q_terms if t in s.lower())
                scored.append((score, s))
            scored.sort(key=lambda x: (-x[0], -len(x[1])))
            best_snips = [s for sc, s in scored[:3] if s]

        # Trim snippets (keep them short)
        trimmed = []
        for sn in best_snips:
            trimmed.append(sn[:400])
        results.append({
            "page": idx + 1,  # 1-based for human readability
            "heading": heading,
            "score": float(sims[idx]),
            "snippets": trimmed
        })
    return results

def detailed_answer_from_pages(query: str, top_k: int = 5) -> str:
    hits = search_pub1075_pages(query, top_k=top_k)
    if not hits:
        return (
            "The app could not access the official PDF at runtime, so detailed citations are unavailable. "
            "Please enable internet access for this Space or try again later. "
            f"Source of truth: {IRS_PUB_1075_URL}"
        )

    out = []
    out.append("### Detailed Guidance (grounded in IRS Publication 1075)")
    # Provide an actionable, structured answer first
    out.append("**Actionable steps:**")
    out.append("- Identify whether the control applies to systems or processes handling Federal Tax Information (FTI).")
    out.append("- Document policy requirements, technical configurations, and operational procedures.")
    out.append("- Implement control mechanisms and verify via monitoring, audits, or tests.")
    out.append("- Maintain evidence (policies, tickets, logs, reports) to demonstrate compliance during reviews.")
    out.append("")

    # Then include the most relevant sections with snippets and exact page numbers
    out.append("**Most relevant sections in Pub. 1075 (by page):**")
    for i, h in enumerate(hits, 1):
        out.append(f"**{i}. Page {h['page']}{h['heading']}**")
        for sn in h["snippets"]:
            out.append(f"> {sn}")
        out.append(f"_Citation: IRS Publication 1075 (official PDF), page {h['page']}. {IRS_PUB_1075_URL}_")
        out.append("")
    # Add a compact reading plan
    pages_list = ", ".join(str(h["page"]) for h in hits[:5])
    out.append(f"**Suggested reading order:** pages {pages_list} in the official PDF above.")
    return "\n".join(out)

# --------------------------------------------------------------------------------------
# Document parsing utils (for uploaded documents)
# --------------------------------------------------------------------------------------
def read_pdf_bytes(file_bytes: bytes) -> str:
    reader = PdfReader(io.BytesIO(file_bytes))
    texts = []
    for page in reader.pages:
        try:
            texts.append(page.extract_text() or "")
        except Exception:
            pass
    return "\n".join(texts)

def read_docx_bytes(file_bytes: bytes) -> str:
    f = io.BytesIO(file_bytes)
    doc = DocxDocument(f)
    return "\n".join([p.text for p in doc.paragraphs])

def extract_text_from_upload(upload_bytes: Optional[bytes]) -> Tuple[str, str]:
    if upload_bytes is None:
        return "", "No file."
    raw = upload_bytes
    # Try PDF
    try:
        txt = read_pdf_bytes(raw)
        if txt.strip():
            return txt, f"PDF file | {len(raw)} bytes | parsed length: {len(txt)} chars"
    except Exception:
        pass
    # Try DOCX
    try:
        txt = read_docx_bytes(raw)
        if txt.strip():
            return txt, f"DOCX file | {len(raw)} bytes | parsed length: {len(txt)} chars"
    except Exception:
        pass
    # Fallback: text
    try:
        txt = raw.decode("utf-8", errors="ignore")
        return txt, f"Plain text | {len(raw)} bytes | parsed length: {len(txt)} chars"
    except Exception as e:
        return "", f"Error reading file: {e}"

def run_checks(doc_text: str) -> List[Dict[str, Any]]:
    results = []
    text = doc_text.lower()
    for chk in CHECKS:
        found = any(re.search(p, text, flags=re.IGNORECASE) for p in chk["patterns_any"])
        status = "Meets (evidence found)" if found else "Gap (no explicit evidence)"
        results.append({
            "title": chk["title"],
            "section": chk["section"],
            "status": status,
            "recommendation": chk["recommendation"]
        })
    return results

def summarize_score(findings: List[Dict[str, Any]]) -> Dict[str, Any]:
    total = len(findings)
    met = sum(1 for f in findings if f["status"].startswith("Meets"))
    gaps = total - met
    score_pct = int(round((met / total) * 100)) if total else 0
    return {"total": total, "met": met, "gaps": gaps, "score": score_pct}

def format_report(meta: str, findings: List[Dict[str, Any]]) -> str:
    summary = summarize_score(findings)
    lines = [
        "# Pub. 1075 Heuristic Compliance Assessment",
        f"- Source of truth: {IRS_PUB_1075_URL}",
        f"- Document: {meta}",
        f"- Summary Score: {summary['score']}% (Met {summary['met']} of {summary['total']}; Gaps {summary['gaps']})",
        "",
        "## Findings (by theme)"
    ]
    for f in findings:
        lines.append(f"### {f['title']}")
        lines.append(f"- Theme: {f['section']}")
        lines.append(f"- Status: {f['status']}")
        lines.append(f"- Recommendation: {f['recommendation']}")
        lines.append("")
    lines += [
        "---",
        "### Notes",
        "- This assessment is heuristic. Controls may be present but phrased differently.",
        "- Validate against the official IRS Publication 1075 and your agency policy."
    ]
    return "\n".join(lines)

# --------------------------------------------------------------------------------------
# Gradio handlers
# --------------------------------------------------------------------------------------
def handle_assessment(upload_bytes: bytes):
    text, meta = extract_text_from_upload(upload_bytes)
    if not text.strip():
        return WARNING_BANNER, "No text extracted. Please upload a PDF or DOCX with selectable text.", "", gr.update(visible=False)

    findings = run_checks(text)
    report_md = format_report(meta, findings)

    report_name = f"pub1075_assessment_{datetime.utcnow().strftime('%Y%m%dT%H%M%SZ')}.md"
    report_path = os.path.join("/tmp", report_name)
    with open(report_path, "wb") as f:
        f.write(report_md.encode("utf-8"))

    table_lines = ["| Control | Status | Theme |", "|---|---|---|"]
    for fnd in findings:
        table_lines.append(f"| {fnd['title']} | {fnd['status']} | {fnd['section']} |")
    table_md = "\n".join(table_lines)

    header_md = (
        f"> {WARNING_BANNER}\n\n"
        f"**Parsed Document Info:** {meta}\n\n"
        f"**Summary Score:** {summarize_score(findings)['score']}%\n\n"
        f"**Authoritative Source:** {IRS_PUB_1075_URL}"
    )
    return header_md, table_md, report_md, report_path

def handle_qa(question: str):
    # Provide a detailed, section-specific answer with page citations
    question = (question or "").strip()
    if not question:
        return "Please enter a question about IRS Publication 1075."
    return detailed_answer_from_pages(question, top_k=5)

# --------------------------------------------------------------------------------------
# UI
# --------------------------------------------------------------------------------------
with gr.Blocks(title=APP_TITLE, theme=gr.themes.Default()) as demo:
    gr.Markdown(f"# {APP_TITLE}\n{APP_TAGLINE}\n\n{WARNING_BANNER}")

    with gr.Tab("Upload & Check (Heuristic)"):
        gr.Markdown(
            "Upload a **PDF** or **DOCX** policy/security document. The assistant will run Pub. 1075-aligned heuristic checks and provide a structured report (downloadable as Markdown)."
        )
        file_in = gr.File(label="Upload PDF or DOCX", file_types=[".pdf", ".docx"], type="binary")
        run_btn = gr.Button("Run Compliance Assessment")
        header_out = gr.Markdown()
        table_out = gr.Markdown()
        report_out = gr.Markdown(label="Full Report (Markdown)")
        download_out = gr.File(label="Download Report (.md)", visible=False)

        run_btn.click(
            fn=handle_assessment,
            inputs=[file_in],
            outputs=[header_out, table_out, report_out, download_out]
        )

    with gr.Tab("Interactive Q&A (Detailed with Page Citations)"):
        gr.Markdown(
            "Ask about Pub. 1075 requirements. The app downloads and searches the **official PDF**, returning detailed guidance "
            "and citing **specific pages** and short snippets.\n\n"
            f"Source of truth: {IRS_PUB_1075_URL}"
        )
        question_in = gr.Textbox(label="Your question", placeholder="e.g., What encryption protections are required for FTI during transmission?")
        ask_btn = gr.Button("Get Answer")
        answer_out = gr.Markdown()
        ask_btn.click(fn=handle_qa, inputs=[question_in], outputs=[answer_out])

    with gr.Tab("About & Scope"):
        gr.Markdown(
            f"""
### Source of Truth
- Only the official IRS Publication 1075 PDF is used: {IRS_PUB_1075_URL}

### How Q&A Works
- The app downloads the PDF (if internet is available), builds a page-level TF-IDF index, and retrieves the most relevant pages.
- It surfaces short, relevant passages and cites **exact page numbers** for deeper reading.

### Security Notes
- Files are processed in memory; the downloadable report is written to **/tmp** solely for user download.
- Do not upload real Federal Tax Information (FTI).

### Limitations
- Heuristic checks may miss controls that are phrased differently.
- This tool does not replace formal IRS compliance review or legal advice.
"""
        )

if __name__ == "__main__":
    # Do not force share=True on Spaces
    demo.launch()