Transcript Mega-Analysis

Feed an entire quarter of meeting transcripts into Claude’s long context window for cross-meeting pattern analysis — recurring objections, shifting priorities, competitor mentions, and sentiment trajectories, then enrich with Mavera

Scenario

Feed an entire quarter of meeting transcripts into Claude’s long context window for cross-meeting pattern analysis. Instead of summarizing each meeting individually, Claude identifies patterns that only emerge across conversations: recurring objections, shifting priorities, competitors mentioned offhand, and sentiment trajectories. Then send the synthesized patterns to Mavera for enrichment with competitive intelligence and trend validation.

Flow: Aggregate transcripts → Anthropic POST /v1/messages (Claude processes all at once) → Extract patterns → Mavera POST /mave/chat → Enriched quarterly intelligence report

Code

Python

import os, glob, time, anthropic, requests

MV = os.environ["MAVERA_API_KEY"]
MV_BASE = "https://app.mavera.io/api/v1"
MV_H = {"Authorization": f"Bearer {MV}", "Content-Type": "application/json"}
client = anthropic.Anthropic()

# 1. Aggregate all transcripts
transcripts = []
for fpath in sorted(glob.glob("./transcripts/2025-Q4/*.txt")):
    with open(fpath) as f:
        transcripts.append(f"--- MEETING: {os.path.basename(fpath)} ---\n{f.read()}")
combined = "\n\n".join(transcripts)
print(f"Loaded {len(transcripts)} transcripts ({len(combined):,} chars, ~{len(combined)//4:,} tokens)")

# 2. Claude cross-meeting pattern analysis
patterns = client.messages.create(
    model="claude-opus-4-6-20250725",
    max_tokens=4096,
    input=[{
        "role": "user",
        "content": "Strategic analyst reviewing an entire quarter of meeting transcripts.\n\n"
            "Identify cross-meeting patterns:\n"
            "1. **Recurring Objections** — concerns across multiple meetings\n"
            "2. **Shifting Priorities** — how priorities evolved early → late quarter\n"
            "3. **Competitor Mentions** — who, how often, in what context\n"
            "4. **Sentiment Arc** — trending more or less optimistic?\n"
            "5. **Decision Bottlenecks** — where deals/projects stall repeatedly\n"
            "6. **Emerging Themes** — topics that appeared late (early signals)\n"
            "7. **Key Stakeholders** — who has most influence\n\n"
            f"Cite specific meetings with verbatim quotes.\n\nTRANSCRIPTS:\n{combined}"
    }],
)
pattern_output = patterns.content[0].text
print(f"Pattern analysis complete — {patterns.usage.input_tokens:,} input tokens")

# 3. Mavera enrichment
enriched = requests.post(f"{MV_BASE}/mave/chat", headers=MV_H, json={
    "message": "Competitive intelligence analyst. Enrich these internal meeting patterns with market context.\n\n"
        f"INTERNAL PATTERNS:\n{pattern_output[:6000]}\n\n"
        "For each pattern:\n"
        "1. **Market Validation** — Do external signals confirm this?\n"
        "2. **Competitive Context** — What are competitors doing about this?\n"
        "3. **Trend Classification** — Short-term blip or structural shift?\n"
        "4. **Recommended Action** — What should leadership do?\n"
        "5. **Risk if Ignored** — What happens if we don't act?\n\n"
        "End with EXECUTIVE SUMMARY (5 bullets, each with urgency: high/medium/low)."
}).json()

print(f"\n{'='*60}\nENRICHED QUARTERLY INTELLIGENCE\n{'='*60}")
print(enriched.get("content", "")[:3000])

Example Output

Loaded 47 transcripts (1,284,920 chars, ~321,230 tokens)
Pattern analysis complete — 321,230 input tokens

ENRICHED QUARTERLY INTELLIGENCE
============================================================
1. RECURRING OBJECTION: "Integration timeline is too long" (12/47 meetings)
   Market Validation: Gartner avg enterprise integration = 4.2 months.
   Competitive Context: Competitor X launched "60-minute setup" in Oct.
   Classification: STRUCTURAL — buyer expectations permanently shifted.
   Risk: HIGH — Pipeline stalls increase 30%/quarter if unaddressed.

2. SHIFTING PRIORITY: Security → compliance framing (early → late quarter)
   Market Validation: SEC cyber disclosure rules took effect Q4.
   Classification: REGULATORY — will intensify through 2026.
   Risk: MEDIUM — Messaging feels dated within 2 quarters.

EXECUTIVE SUMMARY:
• HIGH: Integration speed is the #1 deal-killer — 12/47 meetings
• HIGH: Competitor X gaining mindshare — 8 mentions (up from 2)
• MEDIUM: Compliance framing needed — regulatory shift
• LOW: AI features generating pull — 6 inbound requests

Error Handling

Context window overflow

47 transcripts at ~7K tokens each ≈ 329K tokens — within the 1M limit. If your quarter exceeds 900K tokens, split into months and synthesize in a final call.

Slow responses

Large context calls take 60-120 seconds. The anthropic SDK defaults to 10 minutes. For raw HTTP, set timeout=600.

Mavera context limits

Pattern output can be long. The enrichment call truncates to 6,000 chars. Split into two calls if needed.