Skip to Content
Introduction

Understanding Astrelo: A Deep Dive Into a Production SaaS

I built a SaaS and I’m going to understand every line.

A chapter-by-chapter walkthrough of the Astrelo codebase — a B2B go-to-market intelligence platform with ML scoring, CRM integrations, an AI chat agent, and real-time proactive alerts. Every example is real production code.

What You’ll Learn

Part 1: Foundations

  • Authentication — JWTs, middleware, login flow, the requireAuth pattern
  • The Database — Connection pooling, parameterized queries, migrations, multi-tenancy
  • React Query — QueryClient, query keys, mutations, optimistic updates
  • API Routes — The canonical 10-step handler, validation, error handling

Part 2: The Intelligence Engine

  • Embeddings — 384-dimensional vectors, cosine similarity, NAICS codes
  • Fit Scoring — NAICS matching, size similarity, loss penalties
  • Intent Scoring — Signals, recency decay, semantic topic matching
  • Composite Score — Weighted combination, buying stages, batch scoring
  • Discovery Engine — 80/20 explore/exploit, 7 strategies, segment tracking

Part 3: Integrations

  • HubSpot — OAuth, sync pipeline, entity resolution
  • Salesforce — The alternative CRM path, SOQL, multi-CRM support
  • Slack — Bot tokens, Block Kit, event subscriptions
  • Groq — Two-model LLM architecture, JSON mode, caching

Part 4: Proactive Intelligence

  • Event-Driven Architecture — Webhooks to triggers, the 9-step pipeline
  • Alert Evaluation — 14 triggers, deduplication, rate limiting
  • Predictions — Slip risk, loss risk, champion cold
  • Autonomous Actions — The approval queue, auto-execute, 5 action types
  • AI Content Generation — Async LLM enrichment, structured analysis

Part 5: The Chat Agent

  • Cosmo’s Tool Architecture — Intent classification, tool registry, two-model design
  • Tool Dispatch — Three branches, disambiguation, adaptive enrichment
  • Goldilocks Brain — Strategic recommendations, explore/exploit, segment absorption
  • Action Tools — Email, tasks, CRM sync, the safe vs. confirmation split

Part 6: Production Concerns

  • Error Handling — logError, sanitization, graceful degradation
  • Cron Jobs — 18 scheduled jobs, SKIP LOCKED, timeout budgets
  • Rate Limiting — Fixed-window algorithm, monthly quotas, fail-open design
  • Deployment — Build spec, webpack fallbacks, the monolith advantage

How to Read

Start with Part 1 (Chapters 1-5) in order — they establish patterns you’ll see everywhere. After that, jump to any Part that interests you.

Every code example includes the file path so you can open the actual source and see it in full context.

28 chapters. ~30,000 words. One complete SaaS, explained.

Last updated on
Preface