Supplier exports need cleanup rules that can be checked.
Clean Sheet is my own local Excel cleanup tool and browser report viewer. It trims selected text columns, reports exact duplicate rows and can write a separate workbook from one selected sheet's plain data. The change log links edits and removals back to their original rows. Duplicate removal is an explicit option; the original file stays unchanged.
Need a supplier or catalogue XLSX prepared for import? Message me here on Contra with the full input/output requirements, an anonymised sample and your deadline. I'll review the entire scope and confirm fit, price and timing before starting. I work through written briefs and feedback.
I can handle your Excel tasks, including data entry, cleaning, formatting, merging, and analysis, helping you manage your data efficiently and make better business decisions.
Building a Voice AI Assistant That Turns Phone Calls Into Thoughtful Follow‑Through
I built a Django backend for a family-focused AI phone assistant. A registered caller connects through Twilio, and Django Channels streams the call to an OpenAI Realtime session for a natural, two-way conversation.
The backend adds caller-specific context, selected voices, subscription minute limits, and interruption handling. After the call, background tasks generate a separate summary. A follow-up email is sent only when requested and the recipient is confirmed.
I also built APIs for managing relations, subscriptions, and dashboard usage data. One of the most interesting parts was connecting the whole lifecycle: incoming call, real-time conversation, usage tracking, and post-call follow-through.
Sending the follow-up email only after the recipient is confirmed is a careful rule: a wrong name heard on a call could send it to the wrong person. How does the confirmation work: does the assistant repeat the name during the call, or does the caller confirm later in the dashboard?
Nobody should have to rewatch a 30-minute call to find out what they're supposed to do.
So I built Loom to Tasks. Paste in a meeting transcript and it pulls out the decisions, tasks (with owner and deadline) and blockers in seconds. Python + Gemini with structured outputs.
The rule I cared most about: it never guesses. If nobody said who owns a task, it writes "unknown" instead of inventing a name. Tested against hand-written answer keys and passed 67/67 checks.
Code's on GitHub: github.com/AbrarShuvo/loom-to-tasks
If your team lives in meetings, I can set this up for your workflow.
Writing "unknown" when nobody named an owner is the rule I like most: an invented name is worse than an empty field. My RAG demo follows the same idea and says so when the guide has no answer. In your 67 checks, how many test the "unknown" cases, where the transcript is not clear?