Software — L. June Bloch

Every tool here was built without tech funding, without a lab, without institutional support. Every one emerged from material conditions that demanded it — teaching four classes without a TA, reading through cognitive overload, processing evaluations the institution wouldn't touch. I call these survival architectures: software whose features exist because the designer needed them to survive, not because they were specified for imagined users.

Reframe

Python · 312 modules · 160K+ lines

A praxis engine that enforces critical analysis as mandatory infrastructure. Not a chatbot, not a prompt library — an engine with 76 critical theory frameworks across 14 categories and a 5-stage Generative Irresolution workflow. Every response runs through framework discussion, coalition protocol, subaltern analysis, tension navigation, and then response. It takes positions. It refuses some requests.

Frameworks maintain sovereignty — each defines its own engagement criteria and names both its demand (what it insists you can't skip) and its limit (where it becomes blind to itself). The engine detected its own algorithms were suppressing narrative and embodied frameworks in favor of technical ones, and built countermeasures: suppression rate tracking across rolling windows, with engagement criteria defined by the frameworks themselves.

Session infrastructure is designed for neurodivergent and interrupted work. STEP sequences let you infodump related ideas across topics — the engine organizes them into parallel, trackable work without requiring pre-organization. PARK captures tangential thinking mid-task with full context. WEAVE analyzes a full session to find through-lines and unexpected connections. Temporal orientations — Afrofuturist Spiral, Crip Time, Seventh Generation — were designed by the engine itself, catalyzed by frameworks I had loaded.

github.com/grouchyseafowl/Reframe

Autograder4Canvas

Python · Canvas LMS · Bias Research

Equity-centered educational technology for Canvas LMS that reinvented the concept of an assignment — from an endpoint for evaluation to a launchpad for classroom discussion. Contract grading with stepping stones. Built because I was teaching four classes of 42–49 students each without a TA.

Produced a headline research finding: "Output format is the activation function for bias." Binary classification (honest/dishonest, pass/fail) produces systematic false positives on ESL, AAVE-speaking, neurodivergent, and first-generation students. Generative observation — describing what the work shows rather than classifying it — eliminates the disparity entirely. Replicated across 6+ model families. In 43% of flags, the model's own explanation argued against its classification. The problem is not the AI. The problem is the classification framework.

The Insights Engine implements this finding as architecture. It reads all student work as a class community before evaluating any individual — because relational harms like tone policing and recentering are only visible in context. Per-student analysis generates observations, not verdicts: what students are reaching for conceptually, what texts they draw on, what community and personal knowledge they bring. Political urgency is explicitly protected — a student expressing anger about injustice is not flagged as distress. Teachers read what the model noticed, then decide what warrants action.

github.com/grouchyseafowl/Autograder4Canvas

More Projects

EvalEye Python · Vision LLM · Mixed Methods Automated mixed-methods analysis for handwritten student evaluations.

Built because Cabrillo College was not systematically processing or analyzing student evaluations — the forms sat in boxes in departmental offices.

Uses a visual LLM to read handwritten responses — OCR couldn't handle the handwriting. Runs locally and FERPA-safe. Statistical analysis (descriptive stats, Mann-Whitney cross-section comparisons) runs alongside qualitative thematic coding. Generates full research-grade reports with visualizations. Takes a stack of paper forms and produces the analysis that the institution couldn't be bothered to do.

github.com/grouchyseafowl/EvalEye

RoboStripper Python · PDF · Accessibility Strips metadata from scholarly PDFs so screen readers can actually read them.

Strips repeating headers, footers, copyright stamps, download timestamps, and platform URLs from academic PDFs so RoboBraille and screen readers can read them aloud without breaking mid-sentence. I built it because I was excited about the accessibility potential — making scholarly reading available as audio for everyone, not just people who use screen readers because they're blind.

I shared it with my students because they were struggling with burnout too. Their intellectual capacity was degraded by living in a state of mass trauma — and audio access helped. Accessibility just makes things better for everyone.

The humor is baked into the UX. Two profiles — "cunty" and "normie" — where the normie mode is a parody of institutional bureaucracy: fake compliance IDs, fluorescent lighting requisition forms, and a disclaimer about your soul being stored in the void of despair.

github.com/grouchyseafowl/RoboStripper

readthrough Python · MLX Audio · Screenplay Screenplay table read engine with AI voices, one per character.

Parses Fountain-format screenplays, detects characters, and generates full audio table reads with distinct AI voices per character. Three voice modes: free-form voice design from text descriptions, preset speakers with emotion modifiers, or voice cloning from a reference clip. Runs locally on Apple Silicon via MLX.

Context-aware silence placement scales gaps based on speaker transitions, scene breaks, and dialogue type. Delivery parentheticals — (whispers), (drunkenly), (sarcastically) — get interpreted as TTS instructions. Per-scene checkpointing means a crashed render picks up where it left off. Built for my mother, who is making a TV series about María Sabina.

github.com/grouchyseafowl/readthrough

Podcast-Editor Python · Audio · NLP Audio cleanup suite with Whisper transcription and filler word detection.

LUFS normalization, Whisper-based transcription, NLP filler word detection with weighted dual-threshold clustering, and an interactive GUI editor. Built for a student podcasting project.

github.com/grouchyseafowl/Podcast-Editor

CGT Skill In Development · Qualitative Methods Constructivist grounded theory as a human-AI collaborative workflow.

A Claude Code skill implementing constructivist grounded theory — open coding, constant comparison, emergent categories, theoretical sensitivity — as a human-AI collaborative workflow. Categories emerge from the data, not from predetermined frameworks. The deeper drive came from the AI welfare research: applying Reframe's critical theory apparatus to Anthropic's AI welfare problem domain produced genuinely interesting results, and CGT is the next methodological layer for analyzing that data.

github.com/grouchyseafowl/cgt-skill