Imported from jeremylongshore/tons-of-skills-marketplace (plugins/saas-packs/fathom-pack/skills/fathom-performance-tuning/SKILL.md). Install upstream with npx skills add jeremylongshore/tons-of-skills-marketplace --skill fathom-performance-tuning. Copyright stays with the author (MIT).

Fathom Performance Tuning

Prerequisites

• A baseline for processing latency, sync/delivery error, quality, and an approved data/consent policy.
• Synthetic meeting metadata, an owner, and a reversible performance-change threshold.

Instructions

1. Measure aggregate processing, sync, and alert behavior without using meeting content as diagnostic data.
2. Change one approved queue/concurrency/integration setting and compare against baseline.
3. Revert on quality, consent, delivery, or reliability regression.

Output

• A measured performance recommendation with data/consent guardrails, owner, and rollback record.

Examples

Use synthetic meetings to measure processing and CRM-sync latency, change one bounded concurrency setting, and compare aggregate results. Revert if errors or incorrect follow-up behavior increases; do not bypass review or retention controls to improve performance.

Overview

Fathom's meeting intelligence API serves transcript downloads, bulk meeting sync, and action item aggregation. Transcript payloads are large (50-500KB each), making bulk sync of historical meetings a major latency bottleneck. The 60 req/min rate limit requires careful batching. Caching immutable transcripts aggressively while keeping action item data fresh reduces download latency by 70% and prevents rate limit errors during bulk operations.

Caching Strategy

const cache = new Map<string, { data: any; expiry: number }>();
const TTL = { transcript: 3_600_000, actionItems: 120_000, meetings: 300_000 };

async function cached(key: string, ttlKey: keyof typeof TTL, fn: () => Promise<any>) {
  const entry = cache.get(key);
  if (entry && entry.expiry > Date.now()) return entry.data;
  const data = await fn();
  cache.set(key, { data, expiry: Date.now() + TTL[ttlKey] });
  return data;
}
// Transcripts are immutable — cache 1hr. Action items change — cache 2min.

Batch Operations

async function syncMeetingsBatch(client: any, ids: string[], batchSize = 50) {
  const results = [];
  for (let i = 0; i < ids.length; i += batchSize) {
    const batch = ids.slice(i, i + batchSize);
    const res = await Promise.all(batch.map(id => client.getTranscript(id)));
    results.push(...res);
    if (i + batchSize < ids.length) await new Promise(r => setTimeout(r, 61_000)); // 60 req/min
  }
  return results;
}

Connection Pooling

import { Agent } from 'https';
const agent = new Agent({ keepAlive: true, maxSockets: 6, maxFreeSockets: 3, timeout: 45_000 });
// Transcript downloads are large — longer timeout, fewer concurrent sockets

Rate Limit Management

async function withFathomRateLimit(fn: () => Promise<any>): Promise<any> {
  try { return await fn(); }
  catch (err: any) {
    if (err.status === 429) {
      const retryAfter = parseInt(err.headers?.['retry-after'] || '60') * 1000;
      await new Promise(r => setTimeout(r, retryAfter));
      return fn();
    }
    throw err;
  }
}

Monitoring

const metrics = { downloads: 0, cacheHits: 0, rateLimits: 0, avgLatencyMs: 0 };
function trackDownload(startMs: number, cached: boolean, rateLimited: boolean) {
  metrics.downloads++;
  metrics.avgLatencyMs = (metrics.avgLatencyMs * (metrics.downloads - 1) + (Date.now() - startMs)) / metrics.downloads;
  if (cached) metrics.cacheHits++; if (rateLimited) metrics.rateLimits++;
}

Performance Checklist

• Cache transcripts with 1-hour TTL (immutable after generation)
• Use webhooks instead of polling for new meeting notifications
• Batch transcript downloads in groups of 50 with 60s pauses
• Set action item cache TTL to 2 min for freshness
• Enable HTTP keep-alive with 45s timeout for large payloads
• Track rate limit hits and back off with Retry-After header
• Parallelize independent meeting metadata and transcript fetches
• Aggregate action items client-side to reduce API round-trips

Error Handling

Resources

• Fathom API Docs

Next Steps

See fathom-reference-architecture.