COSMIC-Feedback
About Sign In
This is a live demo of COSMIC-Feedback running with course materials for a simulated class on The Lord of the Rings.

All conversations use real LLM inference.

COSMIC-Feedback

  • Collaborative
  • Socratic
  • Misconception
  • Identification
  • Chatbot

In most classrooms, instructors read the room from faces, tangents, and hallway questions. In a lecture hall of two hundred or a fully asynchronous online section, that feedback isn't there.

COSMIC-Feedback was built for the classes without it, where the instructor can't read the room.

A conversation with every student, before the lecture starts.

A Socratic chatbot talks with every student one-on-one.
It never gives away the answer.
It asks what they think, and then asks why.

The chatbot works from the same readings and lecture the student was assigned. It probes reasoning, requests examples, and follows up on inconsistencies. Students articulate their understanding in their own words, allowing them to clarify their thinking or reveal their confusion.

Simulated pre-class conversation: a real excerpt from an evaluation transcript. Click to explore.

A read of the room, before class starts.

The system analyzes conversations from the entire class and distills what students collectively don't understand.

An analysis pipeline reviews every transcript, clusters recurring misconceptions, and produces a structured briefing. A document to read before class, organized by severity, with excerpts from the actual conversations that surfaced each cluster.

The instructor walks in with real insights into the students' thinking.

Instructor Briefing

Excerpt
July 7, 2026 at 1:58 PM · GPT-OSS 120B · 5 reasoning patterns

15 students · 5 clusters identified · Students reliably link internal values and gradual corruption to Ring resistance, but there are two salient confusions: the bearing vs facing binary and the portrayal of Faramir as a wish‑fulfillment ideal.

Bearing vs. Facing Distinction

Status: Shaky
Prevalence: 8 / 15 ≈ 53%

Students treat the lecture’s bearing‑vs‑facing binary as a rhetorical shortcut that excludes counter‑examples like Faramir, suggesting the categories are fuzzy or artificially imposed.

“If the ‘bearing’ state inherently means corruption is structural, and ‘facing’ means you can resist, then the line between them seems very fuzzy.” – Student A

“The distinction feels like it’s being used to cordon off cases like Faramir, who declines the Ring, which weakens the inevitability thesis.” – Student B

Teaching pivot: When you revisit the binary, pause to map a handful of characters (e.g., Boromir, Sam, Faramir) onto both axes and ask the class to locate any “in‑between” cases. Present the text where Faramir explicitly articulates his reasoning for rejecting the Ring; let students argue whether that line reads as “facing” or a nuanced “bearing” that resists. This concrete mapping will expose the limits of the category and give you a chance to refine it.


Variable Resistance Across Characters

Status: Shaky
Prevalence: 7 / 15 ≈ 47%

Students recognize that exposure length and personal virtues both shape a character’s ability to resist, but they sometimes merge distinct mechanisms (e.g., conflating innate virtue with the amount of time a Ring is borne).

“Galadriel fights a huge internal battle, while Faramir seems to reject the Ring without struggle — different kinds of resistance.” – Student I

“Bilbo’s act of pity at first gave him a stronger shield, but the Ring still eroded him over sixty years.” – Student J

Teaching pivot: Build a quick two‑column visual (Duration vs. Virtue) and plot each major character discussed. Prompt the class to annotate each cell with the textual moment that supports the factor (e.g., Bilbo’s birthday party gift, Galadriel’s Mirror). This will help separate “time‑based erosion” from “value‑based immunity” and clarify where overlap is legitimate versus mistaken.


Faramir as Unrealistic Ideal

Status: Active misconception
Prevalence: 6 / 15 ≈ 40%

Students repeat the lecture’s claim that Faramir is primarily a wish‑fulfillment figure, treating that judgment as fact and downplaying textual evidence of his realistic motivations.

“The lecture frames Faramir as a ‘counter‑example’ that’s more wish‑fulfillment than a believable character.” – Student C

“Faramir feels almost too good to be true — like Tolkien’s ideal of what a person could be, not what a real person would be.” – Student D

Teaching pivot: Introduce a brief close reading of Faramir’s dialogue in The Two Towers (e.g., his reasoning about the Ring’s burden). Ask students to list the concrete concerns he raises (political, moral, practical) and compare them with the lecture’s label. Counter‑balance by showing a scholar who interprets Faramir as a nuanced foil, not a pure ideal. This evidence‑driven contrast should destabilize the blanket “wish‑fulfillment” label.


Internal Values and Mindset

Status: Solid
Prevalence: 9 / 15 ≈ 60%

Students consistently connect a character’s prior moral disposition—humility, love, loyalty—to their capacity to resist the Ring, emphasizing agency over deterministic corruption.

“Faramir’s internal disposition—his wisdom and lack of desire for power—makes the Ring have ‘no purchase’ on him.” – Student E

“Sam’s resistance comes from his unwavering loyalty to Frodo and his simple, unambitious nature.” – Student F


Structural Erosion of Agency Over Time

Status: Solid
Prevalence: 7 / 15 ≈ 47%

Students articulate the lecture’s thesis that prolonged bearing gradually wears down will and identity, describing it as a slow, cumulative disease rather than an instant change.

“If you bear it for a long time, the Ring slowly grinds down your capacity to act on what you know is right.” – Student G

“Frodo’s failure at the Crack of Doom isn’t a personal flaw; it’s the structural truth that the Ring inevitably wears down any bearer.” – Student H

Your data stays yours.

Two distinct LLM roles and an analysis pipeline, all on infrastructure you control.
No student conversation leaves the institution.

The Socratic chatbot and the analysis pipeline both run on self-hosted or institutionally-controlled inference. The models themselves are open source, so nothing here depends on a single vendor. Student names are kept separate from what students actually said. Instructors see the thinking, not who did the thinking.

Runs on your infrastructure

See it for yourself.

This demo runs on course materials for a simulated class on The Lord of the Rings. Two assigned readings, one lecture, and pre-class conversations from an evaluation cohort. Sign in with your Pitt Google account to talk with the chatbot yourself, read the course materials, or view an example instructor briefing generated from a cohort of conversations.

If your Pitt account isn't on the allowlist yet, signing in will send a request and you'll typically be added within a day.