Work Log

I ended up in the ER on 9/6 with walking pneumonia. My O₂ levels were in the low 80s, and it was nearly impossible to keep my eyes open. The brain fog was so severe that I struggled to answer the most basic questions: name, address, date of birth. Thankfully, the staff immediately jumped into action and started pushing IV fluids and antibiotics. By Tuesday morning, 9/9, I was cleared to leave the hospital with a full round of antibiotics and an Albuterol inhaler to keep my airway clear.

All told, I coughed up over 400 ml of mucus from my lungs while in the hospital. Even as I write this, I’m still dealing with random coughing spells that produce more mucus. I’ve coughed so hard that I’ve strained the muscles around my solar plexus.

“That which does not kill us makes us stronger.” ― Friedrich Nietzsche

To help make up for the missed days, I put in solid 10 hour days this week, and will likely do the same next week. Though of course I’ve already been putting in over time since day one when I started. Regardless, I think it’s the honorable thing to try and close the gap and get the projects back on track. I’ve never worked so feverously as I did Tuesday to get my agents back online. Call it pride or honor, but I took it as a personal failure to know they had to take them offline whilst I was out. (Even if my absence was fully justified, and unavoidable.)

Tue 9/9

  • Left the hospital around 10:15 and started organizing discussions to get my agents back online by 11:30.
  • Updated the AI agent server to the latest stable version.
  • Submitted a PR for the counteroffer API.
    • Approved, and agents turned back on!
    • Fixed a JSON bug in the CS Orchestration Agent.
  • Fixed an Amazon bug for v2.0 Returns Agent counteroffers.

Wed 9/10

  • Updated the Warranty Agent to handle highly detailed denial messages with quoted reasoning from the warranty manual.
  • Created an agent to convert HTML → inline plain text format.
    • This will be useful in many workflows.
  • Submitted v2 Return Agent for final approval.
  • Submitted a proposal to leverage vapi.ai to build a blue-ocean warm lead sales associate.
  • Launched v2 Return Agent.
  • Fixed the Amazon Order Number bug (again).
  • Drafted the v2 Returns Agent announcement.
  • Started a Vapi project to handle abandoned carts.

Thu 9/11

  • Fixed a counteroffer bug with the help of our CTO.
  • Reran the failed execution from overnight.
    • Still maintaining a sub-2% failure rate.
    • Not sure I can get below 1% given the APIs I’m hitting and the occasional LLM hallucinations.
  • Submitted a report to the CEO and COO outlining a new abandoned cart policy expansion, and proposed using Vapi to move leads from “warm” to “hot” for CS Reps to close.
  • Pair-coded with our CTO, who identified the root cause of the Amazon Order Number message queue issue by tracing the DB calls and finding where a misformatted email address caused the conflict.
    • My bug fix had been a step in the right direction but didn’t resolve the root problem.
    • Learned a valuable lesson about testing full workflows. Had I done so the first time, I would have at least seen there was still an underlying issue. (Even our CTO admitted it was a tricky one, and he’s a brilliant engineer.)
  • Met with our Shipping Manager to get buy-in on the Vapi AI agent for converting abandoned carts into hot leads.
  • Figured out how to force custom CSS and HTML on RevParts custom pages so our Marketer can create more content-rich “blog post” style articles.
    • Built tools so he can write posts and easily copy them in HTML format to post directly into the site’s special framework.
  • Tested the first steps of Vapi and got the first agent version running.
    • Still need to build out the DB table and access the API endpoint via a cron job (M–F, 9–16).
    • First phone call went well and proved the concept, though there are still plenty of kinks to iron out (like teaching the AI to identify the customer name—LOL).

Fri 9/12

  • Explained my Vapi decisions to the COO and walked through my decision-making process to maximize impact with limited time and resources.
    • Upper management seems to be embracing the idea of giving me more autonomy to pursue projects I believe have the best ROI relative to dev time. We keep using the phrase “low-hanging fruit” as we target the 80/20 projects first. There’s so much room for automation and efficiency gains.
    • Toward this end, I’ve begun compiling a report for my 90-day review, tracking the bottom-line revenue and profit impact of my work. My goal is to quantify my ROI to the company and (hopefully) prove my worth enough to earn a promotion to full-fledged AI Engineer.
  • Began the image project for the WPD site. Looked into the database structure.
    • Pulled down a CSV metadata representation of the S3 buckets.
    • Ran initial tests and saw <10% success.
    • Planning an AI Agent to process and validate the 382,000+ images.
  • Built out three VIN Check APIs:
    • Create the initial queue check.
    • Pull the updated queue for completed VIN Checks.
    • Upsert when completed to prevent duplicate Replyco messaging.
  • Built the AI Agent framework to leverage the new APIs.
    • Still needs extensive testing before rollout to production. And of course, our CTO must sign off on the PRs I’ll submit Monday morning. (Never push on Fridays—LOL).