Selected Work - Product

BUC Populi - a tool that earns faculty trust

An internal tool for a small university, shaped around the people who use it: professors losing hours to grading and an admin team carrying real compliance risk. The product answer wasn't "AI grades essays"-it was "AI drafts, a human always approves."

CRON· weekly automation
live
# crontab 0 7 * * 1 generate_weekly_report.py
MON
07:00report
07:30emails
TUE
08:00refresh
WED
08:00refresh
THU
08:00refresh
FRI
08:00refresh
cron_setup.shnext run · mon 07:00

Role

Solo Engineer / Product

Focus

Pain points · Trust · Time saved

Links

Internal tool

01The product

The users are concrete: professors and a small administration team at one university. Two pains shaped the product. Grading written work eats hours faculty would rather spend teaching, and attendance records have to be defensible for compliance, where a missed week is a real institutional risk. Letting AI grade essays on its own fails the moment you listen to faculty, so the bet flipped: AI does the first pass at scale and the professor stays the system of record, approving every grade.

02What I shipped

  • AI-drafted grades - rubric-anchored essay grading that gives each professor per-criterion scores, feedback, and flags as a starting point-held as a draft, never submitted automatically.

  • Weekly compliance audit - an automated pass that reconciles attendance and submissions per week and flags exactly which students are at attendance-compliance risk - the thing nobody wants to do by hand.

  • A dashboard they already know how to use - a simple web dashboard with no app to install or learn-faculty open a link and work.

  • Hands-off cadence - a weekly cron + email rhythm that delivers the report to the right professor without anyone remembering to run it.

03Product decisions

The human gate is the product

Problem
An AI grade flowing straight into a student's record is unacceptable to faculty, and it would kill adoption on the first day.
Approach
I made the human gate the core feature rather than a setting. Every AI result is a draft at pending_review, nothing is pushed back automatically, and the professor is always the system of record.
Result
Faculty trust the tool enough to use it, because the AI does the volume and the human keeps authority.

Fits the existing workflow

Problem
Professors will not adopt a heavy new app with its own login and learning curve just to save some grading time.
Approach
I shipped a simple web dashboard and email summaries, with no install and no training, in formats they already work in.
Result
The tool fits into the workflow people already have instead of asking them to change it.

Automate the chore, not the call

Problem
The highest-value thing to automate is not the part that needs expertise, it is the repetitive, high-stakes attendance audit nobody wants to own.
Approach
I pointed the automation at the weekly compliance check and the first-pass grading and left the final grade and the academic call to the professor.
Result
The manual weekly attendance checking is largely gone and grading time drops, without anyone giving up control.

Hands-off by default

Problem
A tool busy faculty have to remember to open is the one that gets skipped the week it matters most.
Approach
I made the cadence automatic, so a weekly run delivers the report by email and dashboard with no one starting it.
Result
Adoption does not depend on a new habit, because the work arrives whether or not anyone thinks to ask for it.

04How it was built

Deliberately small and sharp, with the LMS as the source of truth, not a second copy.

Python 3.12Populi API v2Anthropic ClaudeStreamlitSMTPcron

05Outcomes

Weekly

Attendance-compliance check, run by cron

Human-approved

No grade ever auto-pushed