What it does:
Captures leads from 3 sources (webhook, Gmail, Google Sheets), deduplicates, scores with AI via Groq, routes high-score leads to Discord, escalates unresponded leads after 2 hours, and sends a weekly summary report.
3 workflows:
- Main pipeline
- Escalation checker (runs every 30min)
- Error handler
Handled edge cases like spam, invalid emails, non-English messages, missing fields, and duplicate submissions across sources. Built a custom Code node parser after the built-in Structured Output Parser kept failing on double JSON blocks.
The above are examples of high score leads with intent and summary made by the ai llm
Can you please tell me
Anything you’d restructure?
Missing anything for a real client deliverable?
What would you charge for something like this?
4 curtidas
Solid first build — the dedup + escalation + error-handler split is already more production-minded than most “v1” lead systems I see.
A few things I’d restructure:
- Persist the criteria/prompt version alongside each score, not just the score itself. When you tweak the qualification prompt next month, you’ll want to know which leads were scored under the old rules vs the new ones — without it, a strong lead scored “low” under an outdated prompt just looks like a mistake with no way to audit why.
- The 30-min escalation checker polling against a 2hr window is a good pattern, but make sure it’s checking against a persisted “responded” flag rather than re-deriving it from Gmail/Discord state each run — otherwise a slow API response can double-escalate the same lead.
For a real client deliverable, I’d add:
- A confidence field alongside the score, with anything below a threshold routing to manual review instead of auto-scoring. LLMs are confidently wrong often enough that a pure auto-score system needs a “not sure” lane once it’s a client’s actual revenue on the line.
- Test it against deliberately messy input before handing it over — non-English (you’ve got that), but also duplicate-with-slight-variation (same lead, different email) and adversarial-looking submissions. Clients find these edge cases in week one; might as well find them first.
On pricing: for a 3-workflow system like this (main pipeline + escalation + error handling, multi-source ingestion, AI scoring), a reasonable freelance range would be a few hundred to low four figures for setup depending on source complexity, plus a small monthly retainer if you’re maintaining it — the retainer matters more here than the setup fee, since scoring prompts drift and need retuning as the client’s business changes.
