Two-Week Progress Review: Turning Tracker Data into an AI Dashboard | HAI Project Manager

Managing 172 video reviews across 38 fellows.

[0:00]Hi, I'm Sanjana, a project manager at Humanitarian's AI, and this is my two-week progress review, a quick, honest look at the video reviews I ran over the past two weeks. I'll show you how many videos I reviewed, how many went up for the professors to approve, how many needed changes, and the weekly rhythm I keep. Let's use AI to turn my

Establishing an alternating weekly cadence between projects.

[0:20]Tracker into a dashboard a stakeholder can read in a minute. Here's the context. Every week dozens of fellows across many projects submit videos, and someone has to review each one, give clear feedback, and decide what's ready for the professors to see. That someone is me. Over these two weeks that meant 172 videos from 38 fellows. At that volume, a gut feeling of it was a busy couple of weeks isn't enough.

Building a reproducible dashboard using Claude.

[0:45]Stakeholders need the actual numbers, and so do I. Before the numbers, here's the system behind them, because the structure is the point. Two parts. First, a weekly cadence. I alternate every week one week I review the Mycroft project videos. The next week I review all the other projects. That keeps a huge queue from piling up on any single team. Second, every video runs the same pipeline. I review it, then it either

Tracking the quality gate funnel: Approved, needs changes, and blocked.

[1:12]Goes up for the professors to approve, or it goes back with specific changes, or it's blocked on a real issue, and I check whether the code is on GitHub. Same path every time. So, let's build the dashboard with Claude. The ask is simple and specific. Read my tracker, filter to the last two weeks, and give me the totals how many videos I reviewed, how many I sent up for

Verifying GitHub repository compliance across video submissions.

[1:35]Approval, and how many needed changes. Specific in, checkable out. Here's what it wrote. It reads the tracker, keeps only the rows with a video count, and filters to my two-week window. Then two lines do the work. Sum the videos and group them by their approval status. That grouping is the whole report in one move it splits every video into uploaded, needs changes, or blocked. And because this is real code reading my real sheet, every number you're about to see is one I can reproduce. Run it, and here's the headline. 172 videos reviewed. Of those, 119 went straight up for the professors to approve. 50 came back with specific changes, and three were blocked on real issues. Watch the funnel. Most of the work flowed through cleanly, and that 50 isn't waste, it's the quality gate doing its job before anything reaches a professor's desk. But a single total hides my rhythm, and the rhythm is what I actually want stakeholders to see. So, let's revise the ask.

Split the same videos by week and by project, so the alternation shows up one week Micraught, the next everything else. Let's see whether the data matches how I actually plan my weeks. The revision is just two ideas. First, tag each video as Micraught or other. Second, tag it by week. Then group by both. That's it. The comments show what comes back. Week one was other projects, 87 videos. Week two was Micraught 77 with a few stragglers from other teams. The code found the exact cadence I keep without me telling it the answer. Rerun it, and there's the rhythm week by week. Week one was the other projects, 87 videos. Week two was Micraught 85 in total. Two weeks, two focused pushes, alternating exactly as planned. And across the whole fortnight, the split is balanced 95 other project videos and 77 Micraught. This is the picture I want a stakeholder to remember. Steady, alternating, and nothing left to pile up. So, here's the two-week scorecard. Throughput: 172 videos

Reviewed. Quality gate: 119 approved path, 50 sent back for specific fixes, the gate working, not rework wasted, and several fellows have already revised and re-uploaded. Cadence, alternating weeks kept, and compliance, 146 of the 172 had their code on GitHub about 85% with the rest already chased. One honest note, this counts videos reviewed, not hours, and changes requested will keep turning into changes made over the coming days. Your turn, and this one's for the PMs and stakeholders watching. If you keep a review or task tracker, you can get the same dashboard in one prompt. Paste your tracker into Claude and ask for the totals, the split by week and project, your GitHub compliance rate, and three sentences you can drop straight into a stakeholder update. A good answer gives you the cadence and names the blocked items. A bad one just repeats your raw rows back to you. Two-week progress review with Sanjana Rao for Humanitarian's AI. Steady cadence, honest numbers, nothing left to pile up. Thanks for watching.