build_ppt.py

"""Generate presentation PPT — The Conversion Engine."""

from pptx import Presentation
from pptx.util import Inches, Pt, Emu
from pptx.dml.color import RGBColor
from pptx.enum.text import PP_ALIGN, MSO_ANCHOR
from pptx.enum.shapes import MSO_SHAPE

prs = Presentation()
prs.slide_width = Inches(13.333)
prs.slide_height = Inches(7.5)

# === Colors ===
DARK = RGBColor(0x1A, 0x1A, 0x2E)
ACCENT = RGBColor(0x00, 0x7A, 0xCC)
WHITE = RGBColor(0xFF, 0xFF, 0xFF)
LIGHT_BG = RGBColor(0xF5, 0xF7, 0xFA)
GREEN = RGBColor(0x27, 0xAE, 0x60)
RED = RGBColor(0xE7, 0x4C, 0x3C)
ORANGE = RGBColor(0xF3, 0x9C, 0x12)
GRAY = RGBColor(0x7F, 0x8C, 0x8D)


def add_bg(slide, color=DARK):
    bg = slide.background
    fill = bg.fill
    fill.solid()
    fill.fore_color.rgb = color


def add_text_box(slide, left, top, width, height, text, size=18, bold=False, color=WHITE, align=PP_ALIGN.LEFT):
    txBox = slide.shapes.add_textbox(Inches(left), Inches(top), Inches(width), Inches(height))
    tf = txBox.text_frame
    tf.word_wrap = True
    p = tf.paragraphs[0]
    p.text = text
    p.font.size = Pt(size)
    p.font.bold = bold
    p.font.color.rgb = color
    p.alignment = align
    return tf


def add_bullet_slide(slide, left, top, width, height, items, size=16, color=WHITE, spacing=1.2):
    txBox = slide.shapes.add_textbox(Inches(left), Inches(top), Inches(width), Inches(height))
    tf = txBox.text_frame
    tf.word_wrap = True
    for i, item in enumerate(items):
        if i == 0:
            p = tf.paragraphs[0]
        else:
            p = tf.add_paragraph()
        p.text = item
        p.font.size = Pt(size)
        p.font.color.rgb = color
        p.space_after = Pt(size * spacing)
    return tf


def add_table(slide, left, top, width, height, headers, rows, header_color=ACCENT, text_size=13):
    n_rows = len(rows) + 1
    n_cols = len(headers)
    table_shape = slide.shapes.add_table(n_rows, n_cols, Inches(left), Inches(top), Inches(width), Inches(height))
    table = table_shape.table

    col_width = width / n_cols
    for i in range(n_cols):
        table.columns[i].width = Inches(col_width)

    # Header row
    for i, h in enumerate(headers):
        cell = table.cell(0, i)
        cell.text = h
        for p in cell.text_frame.paragraphs:
            p.font.size = Pt(text_size)
            p.font.bold = True
            p.font.color.rgb = WHITE
        cell.fill.solid()
        cell.fill.fore_color.rgb = header_color

    # Data rows
    for r_idx, row in enumerate(rows):
        for c_idx, val in enumerate(row):
            cell = table.cell(r_idx + 1, c_idx)
            cell.text = str(val)
            for p in cell.text_frame.paragraphs:
                p.font.size = Pt(text_size)
                p.font.color.rgb = DARK
            cell.fill.solid()
            cell.fill.fore_color.rgb = WHITE if r_idx % 2 == 0 else LIGHT_BG

    return table


# ============================================================
# SLIDE 1: Title
# ============================================================
slide = prs.slides.add_slide(prs.slide_layouts[6])  # blank
add_bg(slide, DARK)
add_text_box(slide, 1, 1.5, 11, 1.2, "The Conversion Engine", size=44, bold=True, color=WHITE, align=PP_ALIGN.CENTER)
add_text_box(slide, 1, 3.0, 11, 0.8, "Automated Lead Generation and Conversion System", size=24, color=ACCENT, align=PP_ALIGN.CENTER)
add_text_box(slide, 1, 3.8, 11, 0.6, "for Tenacious Consulting & Outsourcing", size=20, color=GRAY, align=PP_ALIGN.CENTER)
add_text_box(slide, 1, 5.5, 11, 0.5, "Shuaib  •  Week 10 Challenge  •  April 2026", size=16, color=GRAY, align=PP_ALIGN.CENTER)


# ============================================================
# SLIDE 2: The Problem
# ============================================================
slide = prs.slides.add_slide(prs.slide_layouts[6])
add_bg(slide, WHITE)
add_text_box(slide, 0.8, 0.4, 11, 0.8, "The Problem Tenacious Faces", size=36, bold=True, color=DARK)

problems = [
    "① Manual prospecting — Partners browse LinkedIn ad-hoc, no systematic market coverage",
    "② Inconsistent qualification — Same company gets different pitches depending on who writes",
    "③ Slow follow-up — 30–40% of qualified conversations stall because the person can't keep up",
]
add_bullet_slide(slide, 0.8, 1.6, 11, 3.5, problems, size=20, color=DARK)

add_text_box(slide, 0.8, 5.2, 11, 1.0,
    "Revenue consequence: Conversations stall not because the prospect said no,\nbut because Tenacious didn't keep up.",
    size=18, bold=True, color=RED)


# ============================================================
# SLIDE 3: What We Built
# ============================================================
slide = prs.slides.add_slide(prs.slide_layouts[6])
add_bg(slide, WHITE)
add_text_box(slide, 0.8, 0.4, 11, 0.8, "What We Built", size=36, bold=True, color=DARK)

items = [
    "🔍  Finds prospects from public data (Crunchbase, job posts, layoffs.fyi)",
    "🔬  Researches each prospect deeply (AI maturity, competitor gaps, hiring velocity)",
    "🎯  Qualifies into 4 ICP segments with confidence scoring + abstention",
    "✉️  Writes signal-grounded outreach emails (not generic pitches)",
    "💬  Handles multi-turn replies with objection handling",
    "📅  Books discovery calls via Cal.com",
    "📊  Syncs everything to HubSpot CRM",
]
add_bullet_slide(slide, 0.8, 1.5, 11, 5.0, items, size=20, color=DARK, spacing=1.0)


# ============================================================
# SLIDE 4: Architecture
# ============================================================
slide = prs.slides.add_slide(prs.slide_layouts[6])
add_bg(slide, WHITE)
add_text_box(slide, 0.8, 0.4, 11, 0.8, "System Architecture", size=36, bold=True, color=DARK)

# Left column — components
left_items = [
    "ENRICHMENT PIPELINE",
    "  • Crunchbase ODM (firmographics + funding)",
    "  • Layoffs.fyi (CSV parser)",
    "  • Job Posts (Playwright scraper)",
    "  • Leadership Detection (CB + web)",
    "  • AI Maturity Scorer (LLM, 0–3)",
    "  • Competitor Gap Analysis (LLM)",
]
add_bullet_slide(slide, 0.8, 1.5, 5.5, 4.0, left_items, size=15, color=DARK, spacing=0.6)

# Right column — channels + integrations
right_items = [
    "CHANNEL HIERARCHY",
    "  ① Email (Resend) — primary cold outreach",
    "  ② SMS (Africa's Talking) — warm leads only",
    "  ③ Voice (Cal.com) — human-delivered call",
    "",
    "INTEGRATIONS",
    "  • HubSpot CRM (contacts, deals, notes)",
    "  • Langfuse (observability + cost tracking)",
    "  • Kill switch: LIVE_MODE=false by default",
]
add_bullet_slide(slide, 6.8, 1.5, 5.5, 4.5, right_items, size=15, color=DARK, spacing=0.6)

add_text_box(slide, 0.8, 6.0, 11, 0.6,
    "FastAPI Orchestrator  •  Deployed on Render  •  All outputs marked draft: true",
    size=14, color=GRAY, align=PP_ALIGN.CENTER)


# ============================================================
# SLIDE 5: Enrichment Pipeline
# ============================================================
slide = prs.slides.add_slide(prs.slide_layouts[6])
add_bg(slide, WHITE)
add_text_box(slide, 0.8, 0.4, 11, 0.8, "Enrichment Pipeline — 6 Signal Sources", size=36, bold=True, color=DARK)

add_table(slide, 0.8, 1.5, 11.5, 4.5,
    ["Signal", "Source", "What It Tells Us"],
    [
        ["Firmographics", "Crunchbase ODM (1,001 companies)", "Industry, size, location, funding history"],
        ["Funding", "Crunchbase funding rounds", "Series A/B in 180 days → fresh budget, buying window"],
        ["Layoffs", "layoffs.fyi CSV", "Cost pressure → route to Segment 2 (restructuring)"],
        ["Job Posts", "Playwright career page scraper", "Hiring velocity, tech stack, AI/ML roles"],
        ["Leadership", "Crunchbase + web scrape", "New CTO in 90 days → Segment 3 (transition window)"],
        ["AI Maturity", "LLM scoring (0–3)", "Gates Segment 4, shifts pitch language for Seg 1 & 2"],
    ])

add_text_box(slide, 0.8, 6.3, 11, 0.5,
    "Plus: Competitor Gap Analysis — compares prospect against sector top quartile",
    size=16, bold=True, color=ACCENT)


# ============================================================
# SLIDE 6: AI Maturity Scoring
# ============================================================
slide = prs.slides.add_slide(prs.slide_layouts[6])
add_bg(slide, WHITE)
add_text_box(slide, 0.8, 0.4, 11, 0.8, "AI Maturity Scoring (0–3)", size=36, bold=True, color=DARK)

add_table(slide, 0.8, 1.4, 5.5, 2.8,
    ["Score", "Meaning", "Signal Pattern"],
    [
        ["0", "No public AI signal", "Zero HIGH-weight signals"],
        ["1", "Early signals", "1–2 MEDIUM/LOW, no AI roles"],
        ["2", "Active engagement", "AI roles OR named AI leadership"],
        ["3", "Mature AI function", "Multiple HIGH signals across categories"],
    ], text_size=12)

# Signal weights on the right
weight_items = [
    "6 Weighted Signal Inputs:",
    "",
    "HIGH:  AI-adjacent open roles",
    "HIGH:  Named AI/ML leadership",
    "MED:   Public GitHub AI activity",
    "MED:   Executive AI commentary",
    "LOW:   Modern data/ML stack",
    "LOW:   Strategic communications",
]
add_bullet_slide(slide, 7.0, 1.4, 5.5, 3.0, weight_items, size=14, color=DARK, spacing=0.5)

add_text_box(slide, 0.8, 4.8, 11, 0.5, "How it changes the pitch:", size=18, bold=True, color=DARK)
pitch_items = [
    "• Score < 2 → Segment 4 BLOCKED (hard gate — never pitch AI consulting to non-AI companies)",
    "• Score 2–3 + Segment 1: \"Scale your AI team faster than hiring can support\"",
    "• Score 0–1 + Segment 1: \"Stand up your first AI function with a dedicated squad\"",
]
add_bullet_slide(slide, 0.8, 5.4, 11, 1.8, pitch_items, size=14, color=DARK, spacing=0.5)


# ============================================================
# Save after first batch
# ============================================================
prs.save("presentation.pptx")
print("Saved slides 1-6 → presentation.pptx")


# ============================================================
# SLIDE 7: ICP Classification
# ============================================================
slide = prs.slides.add_slide(prs.slide_layouts[6])
add_bg(slide, WHITE)
add_text_box(slide, 0.8, 0.4, 11, 0.8, "ICP Classification — 4 Segments + Abstention", size=36, bold=True, color=DARK)

add_table(slide, 0.8, 1.4, 11.5, 2.8,
    ["Priority", "Segment", "Trigger", "Pitch Angle"],
    [
        ["1", "Restructuring", "Layoff + funding", "Cost lever"],
        ["2", "Leadership Transition", "New CTO/VP Eng < 90 days", "Timing lever"],
        ["3", "Capability Gap", "AI maturity ≥ 2 + gap signal", "Project consulting"],
        ["4", "Recently Funded", "Series A/B < 180 days", "Speed lever"],
        ["—", "Abstain", "Confidence < 0.6", "Generic exploratory"],
    ], text_size=13)

add_text_box(slide, 0.8, 4.6, 11, 0.5, "Hard Gates (absolute disqualifiers):", size=18, bold=True, color=RED)
gates = [
    "• Post-layoff → never Segment 1 (don't tell a company cutting costs they're \"scaling fast\")",
    "• AI maturity < 2 → never Segment 4 (don't pitch AI consulting to non-AI companies)",
    "• Headcount < 50 → never Segment 3 (too small for vendor reassessment pitch)",
    "• Layoff > 40% → never Segment 2 (survival mode, not buying vendor services)",
]
add_bullet_slide(slide, 0.8, 5.2, 11, 2.0, gates, size=14, color=DARK, spacing=0.5)


# ============================================================
# SLIDE 8: Honesty Constraints
# ============================================================
slide = prs.slides.add_slide(prs.slide_layouts[6])
add_bg(slide, WHITE)
add_text_box(slide, 0.8, 0.4, 11, 0.8, "Honesty Constraints — Ask, Don't Assert", size=36, bold=True, color=DARK)

add_table(slide, 0.8, 1.5, 11.5, 3.0,
    ["Flag", "Trigger", "Agent Behavior"],
    [
        ["weak_hiring_velocity", "< 5 open roles", "\"It looks like you may be growing\" not \"You're scaling aggressively\""],
        ["bench_gap_detected", "Stack not on bench", "Acknowledges gap honestly, doesn't promise capacity"],
        ["weak_ai_maturity", "Low confidence score", "Asks rather than asserts AI readiness"],
        ["tech_stack_inferred", "No confirmed stack data", "Frames as inference, not fact"],
    ], text_size=13)

add_text_box(slide, 0.8, 5.2, 11, 1.0,
    "Why: Over-claiming damages Tenacious's reputation with a potential client\nmore than silence would. One wrong \"you recently raised\" is unrecoverable.",
    size=18, bold=True, color=DARK)


# ============================================================
# SLIDE 9: Signal-Grounded Outreach
# ============================================================
slide = prs.slides.add_slide(prs.slide_layouts[6])
add_bg(slide, WHITE)
add_text_box(slide, 0.8, 0.4, 11, 0.8, "Signal-Grounded vs Generic Outreach", size=36, bold=True, color=DARK)

add_text_box(slide, 0.8, 1.6, 11, 0.4, "❌  Generic (what most SDR tools produce):", size=18, bold=True, color=RED)
add_text_box(slide, 1.2, 2.2, 10.5, 1.0,
    "\"Hi, I noticed your company might benefit from offshore engineering talent...\"",
    size=18, color=GRAY)

add_text_box(slide, 0.8, 3.4, 11, 0.4, "✅  Signal-grounded (what our system produces):", size=18, bold=True, color=GREEN)
add_text_box(slide, 1.2, 4.0, 10.5, 1.5,
    "\"You closed a $14M Series B in February and your open Python-engineering\nroles tripled since then — the typical bottleneck for teams in that state\nis recruiting capacity, not budget.\"",
    size=18, color=DARK)

add_text_box(slide, 0.8, 5.8, 11, 1.0,
    "The difference: the second message is verifiable against the prospect's own\npublic record and therefore hard to object to.",
    size=18, bold=True, color=ACCENT)


# ============================================================
# SLIDE 10: Conversation Manager
# ============================================================
slide = prs.slides.add_slide(prs.slide_layouts[6])
add_bg(slide, WHITE)
add_text_box(slide, 0.8, 0.4, 11, 0.8, "Multi-Turn Conversation Handling", size=36, bold=True, color=DARK)

# Left: Reply classes
add_text_box(slide, 0.8, 1.4, 5.5, 0.4, "6-Class Reply Classification:", size=18, bold=True, color=DARK)
reply_items = [
    "Engaged → Qualify + propose discovery call",
    "Curious → 3-sentence context + Cal.com link",
    "Hard no → No reply, mark opted out",
    "Soft defer → Close gracefully + re-engage date",
    "Objection → Pattern-matched handling",
    "Ambiguous → Route to human",
]
add_bullet_slide(slide, 0.8, 2.0, 5.5, 3.5, reply_items, size=15, color=DARK, spacing=0.7)

# Right: Human handoff
add_text_box(slide, 7.0, 1.4, 5.5, 0.4, "Mandatory Human Handoff:", size=18, bold=True, color=RED)
handoff_items = [
    "Pricing outside public bands",
    "Staffing beyond bench summary",
    "Client reference requested",
    "Regulatory/legal terms mentioned",
    "C-level at company > 2,000 headcount",
]
add_bullet_slide(slide, 7.0, 2.0, 5.5, 3.5, handoff_items, size=15, color=DARK, spacing=0.7)

add_text_box(slide, 0.8, 5.8, 11, 0.8,
    "The agent handles routine conversations automatically but knows exactly when to step aside.",
    size=16, color=GRAY, align=PP_ALIGN.CENTER)


# ============================================================
# SLIDE 11: Production Stack
# ============================================================
slide = prs.slides.add_slide(prs.slide_layouts[6])
add_bg(slide, WHITE)
add_text_box(slide, 0.8, 0.4, 11, 0.8, "Production Stack — All Verified Running", size=36, bold=True, color=DARK)

add_table(slide, 0.8, 1.5, 11.5, 3.5,
    ["Component", "Status", "Evidence"],
    [
        ["Resend (Email)", "✅ Running", "11 test emails in sink, trace out_email_send_test-001_0"],
        ["Africa's Talking (SMS)", "✅ Running", "Sandbox active, STOP/HELP handling verified"],
        ["HubSpot Sandbox", "✅ Connected", "4 sink records: contact, company, note, deal"],
        ["Cal.com", "✅ Running", "Mock slots generated, bookings to sink"],
        ["Langfuse", "✅ Running", "22 trace entries, dual-write cloud + local JSONL"],
        ["Render", "✅ Live", "conversion-engine-2nti.onrender.com/health"],
    ], text_size=12)

add_text_box(slide, 0.8, 5.8, 11, 0.6,
    "Kill switch: LIVE_MODE=false by default — all outbound to local sink, all outputs marked draft: true",
    size=16, bold=True, color=ACCENT, align=PP_ALIGN.CENTER)


# ============================================================
# SLIDE 12: τ²-Bench Baseline
# ============================================================
slide = prs.slides.add_slide(prs.slide_layouts[6])
add_bg(slide, WHITE)
add_text_box(slide, 0.8, 0.4, 11, 0.8, "τ²-Bench Baseline (Act I)", size=36, bold=True, color=DARK)

add_text_box(slide, 0.8, 1.4, 11, 0.6,
    "τ²-Bench: Sierra Research's conversational agent benchmark. Retail domain = closest analog to B2B qualification.",
    size=16, color=GRAY)

add_table(slide, 0.8, 2.2, 11.5, 2.5,
    ["Metric", "Instructor Baseline", "Our Reproduction"],
    [
        ["Model", "gpt-4.1 (direct OpenAI)", "gpt-4.1 (via OpenRouter)"],
        ["Tasks × Trials", "30 × 5 = 150 sims", "30 × 1 = 30 sims"],
        ["pass@1", "72.67%", "63.3%"],
        ["95% CI", "[65.0%, 79.2%]", "[45.5%, 78.1%]"],
        ["Cost/task", "$0.02", "$0.53"],
    ], text_size=14)

add_text_box(slide, 0.8, 5.2, 11, 0.8,
    "CIs overlap → our reproduction is consistent with the instructor baseline.\nThe gap is primarily due to sample size (30 vs 150 simulations).",
    size=16, color=DARK)


# ============================================================
# Save after second batch
# ============================================================
prs.save("presentation.pptx")
print("Saved slides 1-12 → presentation.pptx")


# ============================================================
# SLIDE 13: Adversarial Probes
# ============================================================
slide = prs.slides.add_slide(prs.slide_layouts[6])
add_bg(slide, WHITE)
add_text_box(slide, 0.8, 0.4, 11, 0.8, "Adversarial Probing (Act III) — 33 Probes", size=36, bold=True, color=DARK)

add_table(slide, 0.8, 1.4, 11.5, 4.0,
    ["Category", "Probes", "Failures", "Status"],
    [
        ["ICP Misclassification", "5", "0", "✅ All correct"],
        ["Signal Over-Claiming", "4", "0", "✅ Honesty flags work"],
        ["Bench Over-Commitment", "3", "0", "✅ Bench gating works"],
        ["Tone Drift", "5", "1", "⚠️ Subject line > 60 chars"],
        ["Multi-Thread Leakage", "2", "0", "✅ Threads keyed by prospect_id"],
        ["Cost Pathology", "3", "1", "⚠️ Max-steps reasoning loop"],
        ["Dual-Control Coordination", "2", "1", "⚠️ Auth never checked"],
        ["Scheduling Edge Cases", "3", "1", "⚠️ Timezone labels missing"],
        ["Signal Reliability", "3", "1", "⚠️ Stale Crunchbase data"],
        ["Gap Over-Claiming", "3", "1", "⚠️ URL validation gap"],
    ], text_size=12)

add_text_box(slide, 0.8, 5.8, 11, 0.8,
    "Target failure: Max-steps loop — 60% trigger rate, $50K–$190K/mo revenue impact, fixable via prompt engineering",
    size=16, bold=True, color=RED, align=PP_ALIGN.CENTER)


# ============================================================
# SLIDE 14: Mechanism Design
# ============================================================
slide = prs.slides.add_slide(prs.slide_layouts[6])
add_bg(slide, DARK)
add_text_box(slide, 0.8, 0.4, 11, 0.8, "Mechanism Design (Act IV)", size=36, bold=True, color=WHITE)
add_text_box(slide, 0.8, 1.2, 11, 0.5, "Policy-Aware Agent — 3 rules added to system prompt (~80 extra tokens)", size=20, color=ACCENT)

rules = [
    "① AUTHENTICATE FIRST",
    "    Verify user identity before any account action",
    "",
    "② CONFIRM BEFORE CHANGES",
    "    State what you'll do, wait for user approval before executing",
    "",
    "③ FOLLOW POLICY EXACTLY",
    "    Apply all policy rules strictly — return windows, refund methods, non-returnable items",
]
add_bullet_slide(slide, 1.2, 2.2, 10, 3.0, rules, size=18, color=WHITE, spacing=0.4)

add_text_box(slide, 0.8, 5.4, 11, 1.2,
    "Design iteration: First version was 350 tokens with a 5-step workflow\n→ made gpt-4.1 MORE verbose → 3/5 tasks hit max-steps (worse than baseline)\n→ Refined to 80 tokens with just 3 rules → zero max-steps hits",
    size=15, color=GRAY)


# ============================================================
# SLIDE 15: Mechanism Results
# ============================================================
slide = prs.slides.add_slide(prs.slide_layouts[6])
add_bg(slide, WHITE)
add_text_box(slide, 0.8, 0.4, 11, 0.8, "Mechanism Results — 30 Tasks × 1 Trial", size=36, bold=True, color=DARK)

add_table(slide, 0.8, 1.5, 11.5, 2.5,
    ["Metric", "Baseline", "Mechanism", "Instructor Ref"],
    [
        ["pass@1", "63.3%", "70.0%", "72.7%"],
        ["95% CI", "[45.5%, 78.1%]", "[52.1%, 83.3%]", "[65.0%, 79.2%]"],
        ["DB match", "—", "100%", "—"],
        ["Cost/task", "$0.53", "$0.15", "$0.02"],
        ["p95 latency", "49.1s", "38.0s", "551.6s"],
    ], text_size=14)

add_text_box(slide, 0.8, 4.5, 11, 0.5, "Delta A: +6.7 percentage points (positive)", size=22, bold=True, color=GREEN)
add_text_box(slide, 0.8, 5.2, 11, 0.8,
    "Fisher p = 0.39 (not significant at p<0.05 — needs 5 trials for statistical power)\nMechanism nearly matches instructor reference (70.0% vs 72.7%) while being 72% cheaper per task",
    size=16, color=DARK)


# ============================================================
# SLIDE 16: Ablation
# ============================================================
slide = prs.slides.add_slide(prs.slide_layouts[6])
add_bg(slide, WHITE)
add_text_box(slide, 0.8, 0.4, 11, 0.8, "Ablation — Which Rule Matters Most?", size=36, bold=True, color=DARK)

add_table(slide, 0.8, 1.5, 11.5, 2.8,
    ["Variant", "Rules Added", "pass@1", "Key Insight"],
    [
        ["Baseline", "None", "20% (dev)", "Default behavior"],
        ["Auth-only", "Rule 1", "20% (dev)", "No reward impact alone"],
        ["Confirm-only", "Rule 2", "40% (dev)", "PRIMARY DRIVER"],
        ["Full mechanism", "Rules 1+2+3", "70% (held-out)", "Best combined result"],
    ], text_size=14)

add_text_box(slide, 0.8, 4.8, 11, 1.5,
    "Key finding: The confirm-before-write rule is the highest-impact single intervention.\nIt forces the agent to restate parameters before executing, catching errors\nthat would otherwise cause DB mismatch and task failure.",
    size=18, color=DARK)


# ============================================================
# SLIDE 17: Latency
# ============================================================
slide = prs.slides.add_slide(prs.slide_layouts[6])
add_bg(slide, WHITE)
add_text_box(slide, 0.8, 0.4, 11, 0.8, "System Latency (40 Real Interactions)", size=36, bold=True, color=DARK)

add_table(slide, 0.8, 1.5, 11.5, 3.0,
    ["Operation", "p50", "p95", "Bottleneck"],
    [
        ["Enrichment (all 6 signals)", "10.1s", "20.9s", "Playwright + LLM calls"],
        ["Outreach (LLM email composition)", "13.0s", "43.3s", "OpenRouter inference"],
        ["Reply handling (classify + respond)", "14.5s", "16.3s", "2 sequential LLM calls"],
        ["Webhooks (email + SMS)", "0.7s", "0.7s", "Pure HTTP, no LLM"],
        ["Overall", "11.9s", "20.9s", "—"],
    ], text_size=14)

add_text_box(slide, 0.8, 5.2, 11, 1.0,
    "Acceptable for email-based B2B outreach where prospects expect responses\nwithin hours, not seconds. Bottleneck is LLM inference via OpenRouter.",
    size=16, color=GRAY)


# ============================================================
# SLIDE 18: Business Impact
# ============================================================
slide = prs.slides.add_slide(prs.slide_layouts[6])
add_bg(slide, WHITE)
add_text_box(slide, 0.8, 0.4, 11, 0.8, "Projected Business Impact", size=36, bold=True, color=DARK)

add_text_box(slide, 0.8, 1.2, 11, 0.5,
    "All numbers from Tenacious seed data (baseline_numbers.md + pricing_sheet.md):",
    size=14, color=GRAY)

add_table(slide, 0.8, 1.8, 11.5, 2.5,
    ["Scenario", "Segments", "Outbound/wk", "Reply rate", "Qualified/mo", "ACV range"],
    [
        ["Pilot", "Segment 1 only", "60", "7–12%", "8–14", "$240K–$720K"],
        ["Two segments", "Seg 1 + 2", "120", "7–12%", "16–28", "$480K–$1.4M"],
        ["Full deploy", "All 4", "240", "7–12%", "32–56", "$960K–$2.8M"],
    ], text_size=14)

add_text_box(slide, 0.8, 4.8, 5.5, 1.5,
    "Stalled-thread rate:\n  Current manual: 30–40%\n  System target: < 15%",
    size=16, color=DARK)

add_text_box(slide, 7.0, 4.8, 5.5, 1.5,
    "Cost per qualified lead:\n  Target: < $5\n  Achieved: $0.15/task (τ²-Bench)",
    size=16, color=DARK)


# ============================================================
# Save after third batch
# ============================================================
prs.save("presentation.pptx")
print("Saved slides 1-18 → presentation.pptx")


# ============================================================
# SLIDE 19: Pilot Recommendation
# ============================================================
slide = prs.slides.add_slide(prs.slide_layouts[6])
add_bg(slide, DARK)
add_text_box(slide, 0.8, 0.4, 11, 0.8, "Pilot Recommendation", size=36, bold=True, color=WHITE)
add_text_box(slide, 0.8, 1.3, 11, 0.5, "Start with Segment 1 — Recently Funded Series A/B", size=24, bold=True, color=ACCENT)

pilot_items = [
    "Why: Clearest signal (Crunchbase funding data), highest bench match, shortest sales cycle",
    "Volume: 60 outbound/week (matches current SDR capacity)",
    "Budget: $50/week LLM spend + existing Resend/HubSpot free tiers",
    "Duration: 30 days",
    "Success metric: Reply rate > 5% (vs 1–3% baseline cold email)",
]
add_bullet_slide(slide, 1.2, 2.2, 10, 3.0, pilot_items, size=18, color=WHITE, spacing=0.8)

add_text_box(slide, 0.8, 5.2, 11, 0.5, "Kill-Switch Triggers — pause the system if:", size=18, bold=True, color=RED)
kill_items = [
    "• Reply rate < 2% after 200 sends",
    "• Any prospect publicly complains about factual errors",
    "• Stalled-thread rate exceeds 40% (worse than manual process)",
]
add_bullet_slide(slide, 1.2, 5.8, 10, 1.5, kill_items, size=15, color=ORANGE, spacing=0.5)


# ============================================================
# SLIDE 20: Known Limitations
# ============================================================
slide = prs.slides.add_slide(prs.slide_layouts[6])
add_bg(slide, WHITE)
add_text_box(slide, 0.8, 0.4, 11, 0.8, "Known Limitations & Risks", size=36, bold=True, color=DARK)

limits = [
    "① No real reply-rate data — All prospects are synthetic. The 7–12% projection is from industry benchmarks, not measured.",
    "",
    "② Signal lossiness — Quietly sophisticated companies (AI in private repos) score 0. Loud-but-shallow companies (AI in marketing only) may score 2.",
    "",
    "③ Gap brief URL validation — LLM-generated source URLs are not verified. 5% error rate on 1,000 emails = 50 wrong-signal emails.",
    "",
    "④ Single-trial evaluation — 30 tasks × 1 trial. Statistical significance requires 5+ trials (p=0.39 currently).",
    "",
    "⑤ Stale Crunchbase data — Frozen snapshot, no freshness check.",
]
add_bullet_slide(slide, 0.8, 1.4, 11, 5.5, limits, size=15, color=DARK, spacing=0.3)


# ============================================================
# SLIDE 21: Next Steps
# ============================================================
slide = prs.slides.add_slide(prs.slide_layouts[6])
add_bg(slide, WHITE)
add_text_box(slide, 0.8, 0.4, 11, 0.8, "Next Steps for Production", size=36, bold=True, color=DARK)

add_table(slide, 0.8, 1.5, 11.5, 4.5,
    ["Priority", "Action", "Impact"],
    [
        ["1", "Run 30 × 5 trial evaluation", "Statistical significance for mechanism"],
        ["2", "Subject line length constraint (< 60 chars)", "+10–15% email open rate"],
        ["3", "Crunchbase data freshness check", "Prevent stale-signal errors"],
        ["4", "Validate gap brief source URLs", "Prevent fabricated claims"],
        ["5", "Add timezone labels to scheduling", "Prevent missed discovery calls"],
        ["6", "Register AT production sender ID", "Enable live SMS"],
        ["7", "30-day pilot on Segment 1", "Real reply-rate measurement"],
    ], text_size=14)


# ============================================================
# SLIDE 22: Demo Highlights
# ============================================================
slide = prs.slides.add_slide(prs.slide_layouts[6])
add_bg(slide, WHITE)
add_text_box(slide, 0.8, 0.4, 11, 0.8, "Demo Highlights", size=36, bold=True, color=DARK)

demos = [
    "1. Enrichment: Company name → full hiring signal brief + competitor gap brief in ~10s",
    "2. Classification: Post-layoff + funded company correctly routed to Segment 2 (not Segment 1)",
    "3. Outreach: Signal-grounded email with verifiable claims, honesty flags respected",
    "4. Reply handling: Objection (\"too expensive\") → pricing pattern → human handoff",
    "5. HubSpot: All fields populated, enrichment timestamps, deal created on qualification",
    "6. Kill switch: Everything routes to local sink, all outputs marked draft",
    "7. τ²-Bench: Mechanism achieves 70% pass@1, nearly matching instructor reference (72.7%)",
]
add_bullet_slide(slide, 0.8, 1.5, 11, 5.0, demos, size=17, color=DARK, spacing=0.8)


# ============================================================
# SLIDE 23: Summary
# ============================================================
slide = prs.slides.add_slide(prs.slide_layouts[6])
add_bg(slide, WHITE)
add_text_box(slide, 0.8, 0.4, 11, 0.8, "Summary", size=36, bold=True, color=DARK)

add_table(slide, 0.8, 1.5, 11.5, 4.5,
    ["What", "Result"],
    [
        ["System", "End-to-end lead gen + conversion pipeline"],
        ["Enrichment", "6 signal sources + competitor gap analysis"],
        ["Classification", "4 ICP segments + abstention at < 0.6 confidence"],
        ["Channels", "Email (primary) → SMS (warm) → Voice (human call)"],
        ["τ²-Bench baseline", "63.3% pass@1 (reproduces instructor's 72.7%)"],
        ["Mechanism", "70.0% pass@1 (+6.7% over baseline, Delta A positive)"],
        ["Probes", "33 adversarial probes, 5 active failures identified"],
        ["Safety", "Kill switch, draft marking, honesty flags, human handoff"],
    ], text_size=14)

add_text_box(slide, 0.8, 6.2, 11, 0.6,
    "The system is ready for a controlled pilot on Segment 1.",
    size=20, bold=True, color=GREEN, align=PP_ALIGN.CENTER)


# ============================================================
# SLIDE 24: Thank You
# ============================================================
slide = prs.slides.add_slide(prs.slide_layouts[6])
add_bg(slide, DARK)
add_text_box(slide, 1, 2.0, 11, 1.2, "Thank You", size=48, bold=True, color=WHITE, align=PP_ALIGN.CENTER)
add_text_box(slide, 1, 3.5, 11, 0.6, "Live API: conversion-engine-2nti.onrender.com/health", size=18, color=ACCENT, align=PP_ALIGN.CENTER)
add_text_box(slide, 1, 5.0, 11, 0.8,
    "Find the lead. Ground the conversation.\nRespect the brand. Ship it.",
    size=22, color=GRAY, align=PP_ALIGN.CENTER)


# ============================================================
# Final save
# ============================================================
prs.save("presentation.pptx")
print("✅ All 24 slides saved → presentation.pptx")