How to Turn Meeting Transcripts into Agent Memory

import os, requests, json, re

API_KEY = os.environ['SCAVIO_API_KEY']
SH = {'x-api-key': API_KEY, 'Content-Type': 'application/json'}

def parse_transcript(text):
    """Extract structured data from meeting transcript."""
    memory = {'decisions': [], 'action_items': [], 'topics': [], 'references': []}
    lines = text.split('\n')
    for line in lines:
        line_lower = line.lower().strip()
        if any(w in line_lower for w in ['decided', 'agreed', 'approved', 'confirmed']):
            memory['decisions'].append(line.strip())
        if any(w in line_lower for w in ['will do', 'action:', 'todo:', 'assigned to', 'by friday', 'by next']):
            memory['action_items'].append(line.strip())
        urls = re.findall(r'https?://[^\s]+', line)
        memory['references'].extend(urls)
        # Extract mentioned tools/companies
        for word in line.split():
            if word[0:1].isupper() and len(word) > 3 and word.isalpha():
                if word not in ['The', 'This', 'That', 'They', 'What', 'When', 'Where']:
                    memory['topics'].append(word)
    memory['topics'] = list(set(memory['topics']))[:10]
    memory['references'] = list(set(memory['references']))
    return memory

# Test with sample transcript
sample = """John: We decided to switch from Tavily to a cheaper search API.
Sarah: Agreed. I will do the migration by next Friday.
John: Let's use Scavio. See https://scavio.dev for docs.
Sarah: Action: update the n8n workflow to use new API.
John: Confirmed the budget is $50/month for search."""

memory = parse_transcript(sample)
for key, items in memory.items():
    print(f'{key}: {items}')

def enrich_memory(memory):
    """Add search context for mentioned topics and references."""
    enriched = []
    # Enrich topics with current info
    for topic in memory.get('topics', [])[:5]:
        data = requests.post('https://api.scavio.dev/api/v1/search',
            headers=SH, json={'query': f'{topic} latest 2026', 'country_code': 'us', 'num_results': 2}).json()
        results = data.get('organic_results', [])
        if results:
            enriched.append({
                'topic': topic,
                'context': results[0].get('snippet', '')[:150],
                'url': results[0].get('link', '')
            })
    memory['enriched_topics'] = enriched
    print(f'Enriched {len(enriched)} topics')
    for e in enriched:
        print(f'  {e["topic"]}: {e["context"][:60]}...')
    print(f'Enrichment cost: ${len(enriched) * 0.005:.3f}')
    return memory

memory = enrich_memory(memory)
print(f'\nFull memory keys: {list(memory.keys())}')

def save_memory(memory, meeting_date='2026-05-20', filename='meeting_memory.json'):
    """Save structured meeting memory for agent consumption."""
    record = {
        'date': meeting_date,
        'type': 'meeting_transcript',
        'decisions': memory.get('decisions', []),
        'action_items': memory.get('action_items', []),
        'topics': memory.get('topics', []),
        'enriched_context': memory.get('enriched_topics', []),
        'references': memory.get('references', []),
    }
    with open(filename, 'w') as f:
        json.dump(record, f, indent=2)
    print(f'Meeting memory saved to {filename}')
    print(f'\n=== Memory Summary ===')
    print(f'  Date: {record["date"]}')
    print(f'  Decisions: {len(record["decisions"])}')
    print(f'  Action items: {len(record["action_items"])}')
    print(f'  Topics: {len(record["topics"])}')
    print(f'  Enriched context: {len(record["enriched_context"])}')
    print(f'\n  Decisions:')
    for d in record['decisions']:
        print(f'    - {d[:60]}')
    print(f'  Action items:')
    for a in record['action_items']:
        print(f'    - {a[:60]}')
    return record

save_memory(memory)