Instructor evaluation, on every class instead of a sample
Circuit Stream runs immersive technology summer camps on six university campuses across the US and Canada. Evaluating instructors meant a coordinator sitting through a class, so in practice they sampled one session per instructor and hoped it was representative. We built them a system that scores every class against their own four-category rubric and returns a finished report in about half an hour.
Live Session Analytics
Introduction to Computer Science • Prof. Johnson • 24 students
Student Attention Map
Emotion Analysis(46 dimensions)
Voice AI Tone Analysis
Enthusiastic
Teacher Evaluation(1-4 scale)
Speaking Time Distribution
Session Timeline
Active Alerts
3 students below attention threshold
Seats 4, 8, 13
Comprehension drop detected at 32:15
Topic: Recursion
- Scored, where before it was one sampled session per instructor
- Every classScored, where before it was one sampled session per instructor
- In the US and Canada, three of them uploading classes in the pilot
- 6 campusesIn the US and Canada, three of them uploading classes in the pilot
- From AI and machine learning to medicine, all on one rubric
- 4 programmesFrom AI and machine learning to medicine, all on one rubric
- Less coordinator time spent evaluating instructors
- 79%Less coordinator time spent evaluating instructors
- From upload to a finished report
- 27 minFrom upload to a finished report
This is not the same evaluation done faster. Evaluating every class, from every instructor, was never possible before.
- A scored evaluation on the client's own four-category rubric
- Strengths, improvements and a written action plan
- Every student question, timestamped
- Planned curriculum against what was actually taught
- Voice signals: concentration, confidence, room noise
- A jump from any finding straight to that moment
Across classes the dashboard names who has not recorded, who dropped a full performance band, who swings between sessions, and how every instructor, course and campus ranks against the rest.
79% less coordinator time
The median class runs 1 h 10 m and nobody has to sit through it: the report is ready in about 40 minutes, median 27, unattended. A fifteen-minute read replaces a seventy-minute class, and the coaching happens the next morning rather than three weeks later.
2,058 statements scored
Every class judged against the same four categories. Two coordinators drift apart from each other; one coordinator drifts across a six-hour day. The rubric does neither.
4 programmes, 3 campuses
Classes arrived from three campuses at once, in four programmes that have nothing in common but the rubric they are judged against, and one of them is medicine while another is video games. Nobody was going to sit through all of them, so in practice you sample one session per instructor and hope it was representative. Every class was analysed instead.
8 hours in 60 seconds
On 3 July, eight hours of class were uploaded inside a 60-second window. A person needs a full working day plus overtime to clear that. Both reports came out.
- Rubric judgements, scored 1 to 4
- 2,058Rubric judgements, scored 1 to 4
- Written coaching items
- 1,542Written coaching items
- Key moments flagged
- 684Key moments flagged
- Transcript lines, speaker-separated
- 33,428Transcript lines, speaker-separated
- Student questions captured
- 827Student questions captured
- Themes identified across classes
- 666Themes identified across classes
The rubric discriminates instead of rubber-stamping
Scores spread rather than cluster: 43% of all judgements landed below Proficient: 10.6% at Needs Improvement, 32.6% at Developing. An evaluation that scores everyone the same is not an evaluation.
Planned curriculum against what was actually taught
Where a lesson plan was supplied, which in this pilot was four classes, 60 of the 64 planned topics were covered, and the report names the four that were not, with the timestamp.
“Instead of reviewing hours of recordings, I can quickly identify and act on the highest-impact coaching opportunities.”
“Basetone.AI is an important part of my OKRs, ensuring that I have a strong understanding of the platform, establish continuity in how we use it, and ultimately maximize the value we get from the software and partnership.”
FinanceHawkclaw Capital Advisors
Family office · Switzerland
An earnings desk that covers the whole portfolio
Earnings season arrives for the whole portfolio at once. The desk reads every call, including the ones there was never time for.
See what we builtGlobal investment bank
The questions a founder steers around
A global investment bank, management meetings from a live deal process, and everything the room noticed but nobody wrote down.
See what we builtDefence, intelligence and law enforcement
Interviews reviewed line by line, including the ones nobody had time for
Contradictions across sessions, questions answered around, delivery against the speaker's own baseline, accent and dialect markers, and what the video shows at the same moment.
See what we builtTell us the decision you want to fix
One call, no deck. We will tell you whether this is something we can build for you, and if it is not, we will say that too.