🪴 Brain

Tech To The Rescue

3 min read

Tech To The Rescue

🗒️ Description

A PLSoft advisory engagement with a tech-for-good nonprofit that connects tech companies and volunteers with NGOs. The work is a strategic IT diagnosis and enablement: move the organization from a fragmented no-code/SaaS stack toward an AI-native, unified platform — consolidating data sources, embedding AI into operations, and turning ad-hoc automation into governed, auditable skills.

🧩 Scope

  • Audit the existing data architecture and integration patterns
  • Define the operating model (who owns which layer of the stack)
  • Design an adoption path for AI (skills framework, not just more tools)
  • Identify quick wins vs. structural changes

Technology (platform under review)

Stack

⚙️ Diagnosis know-how (reusable)

  • The source-of-truth conflict is organizational, not technical. A well-designed Postgres schema can still be bypassed if the operational source of truth lives in an ops tool (Monday) where records are created in conversations and never reach the database. Map where data is actually born before designing sync.
  • Asymmetric bidirectional sync is fragile. DB↔ops-tool sync via queue → webhook → automation tool tends to grow infinite-loop workarounds, miss certain field-change events, and lack a staging environment. Daily failure alerts get ignored once they’re routine — treat alert fatigue as a design smell.
  • Schema debt from no-code migrations. Zero-copy migrations off Airtable carry 150–200 legacy columns and duplicate status fields. New, fast-changing forms are better stored as a single JSON metadata blob to avoid weekly migrations.
  • Embedding quality depends on chunking, not just the model. One embedding per entity over full body text dilutes keyword matches and spreads similarity scores unpredictably — scope and chunk before indexing.
  • Shadow IT is an asset to formalize, not forbid. Effective ad-hoc AI tools built by non-engineers should be promoted into governed, owned, audited skills (Slack/MCP-fronted agents reading the database + docs) rather than banned.
  • 70-20-10: ~70% of AI value is people + process, not technology. Adoption routines beat tool count; training that doesn’t form habits doesn’t stick.

🔗 Links

Graph View

Backlinks