Description
Summary I run a financial services firm. I need an automation agent that processes my daily communications and stages CRM updates for my approval. Clear spec below. I want a builder who has shipped LLM + Notion automation projects before, not someone learning on my dime. I am not prescribing the stack. n8n is one option, but propose whatever gets this done in the simplest, most reliable way. I care about three things: it works every day without babysitting, I own everything, and it can be trained and improved over time as I correct its outputs. What it does: Runs on a daily schedule (6pm ET) plus a manual run button Pulls all Gmail threads with activity in the last 24 hours Pulls all Fireflies.ai call transcripts from the same window Matches each email and call to existing leads across multiple Notion databases (by email address, company name, phone). Unmatched items get proposed as new rows Uses an LLM (Claude API preferred) with a business context prompt I will provide to generate, per touchpoint: a summary note, suggested field updates (status, follow-up date), target database, target row, and a draft follow-up email when warranted Writes every proposal to a staging area with Pending status. Nothing touches my real databases without approval Review interface: a simple dashboard, one card per proposal, where I can edit the note text, change the target database and row via dropdowns, edit the draft email, then approve or reject per card or approve all - ideally a notion database with status column for approval On approval: the note and field updates write to the correct Notion database, follow-up emails are created as Gmail drafts (never auto sent), follow-up dates pushed to Google Calendar Daily notification when the batch is ready If a day is missed, the next run picks up everything since the last successful batch Trainability requirement: Should be able to add rules as I use. My edits and rejections should be captured and usable to improve routing and writing quality over time, whether through prompt refinement, few-shot examples pulled from approved outputs, or a feedback log I can review with you. Propose your approach. I do not want a static build that produces the same mistakes in month six that it made in week one. SGS requirements: Your choice of orchestration: n8n, Make, custom code, or otherwise. Justify it. Simplest and most maintainable wins Claude API (Anthropic) preferred for the LLM layer Notion API, Gmail API, Google Calendar API, Fireflies API Dashboard can be lightweight. Propose your approach All credentials and infrastructure in my accounts. I own everything at handoff, including the ability to modify prompts myself Deliverables: Working end to end system Documentation: architecture overview, how to modify the prompt, how to add a new Notion database to the router, how the feedback loop works 4 weeks of bug fix support post delivery To apply, answer these: Link one project you built involving an LLM API and Notion or a similar CRM How would you handle matching an email to the right lead when the same contact exists in two databases? How would you make this system improve from my corrections over time? Be specific Your fixed price and timeline for this exact scope Fixed price only. No hourly.