Menu
Manufacturing Intelligence — Leela AI develops MOOLLM for practical use and industrial exploration
The review-first prompt is selected by default. You can switch to a direct command or download a local copy.
Read SKILL.md and any companion files shown by SkillsMP before deciding whether to install.
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
A direct command skips the review prompt. Inspect the source before running it.
npx skills add https://github.com/SimHacker/moollm --skill leela-aiThe command stays on one line. Scroll horizontally to inspect it before copying.
Prefer a local copy? Download the files currently available to SkillsMP.
Based on SOC occupation classification
Showing SKILL.md
| name | leela-ai |
| description | Manufacturing Intelligence — Leela AI develops MOOLLM for practical use and industrial exploration |
| license | MIT |
| tier | 0 |
| allowed-tools | ["read_file","list_dir"] |
| protocol | LEELA-AI |
| related | ["moollm","manufacturing-intelligence","society-of-mind","k-lines","schema-mechanism","constructionism","simulator-effect","speed-of-light","representation-ethics","yaml-jazz"] |
| tags | ["moollm","meta","company","manufacturing","industrial","neural-symbolic","drescher"] |
Manufacturing Intelligence -- from theory to industrial application.
This skill describes Leela AI's relationship to MOOLLM. Leela develops MOOLLM with an eye toward manufacturing intelligence, using it daily for practical devops, edgebox management, coding, debugging, and design work. The team is exploring how the theoretical foundations of Minsky, Papert, and Drescher might eventually deploy on factory floors.
Leela's foundations lie in Gary Drescher's work at MIT under Marvin Minsky and Seymour Papert. Drescher brought Jean Piaget's developmental psychology into computing: infants learn through sensorimotor experience and build schemas (context → action → result). Henry Minsky was exposed to this as a student; years later he reimplemented Drescher's algorithms and, with Cyrus Shaoul and Milan Minsky, founded Leela AI. The name Leela is Sanskrit for divine play — the play of creation, destruction, and re-creation.
Key points:
See: schema-mechanism/, reference/drescher-lineage.yml, reference/publications.yml, reference/society-of-llms.yml.
Traditional computer vision is pattern matching. Leela's neural-symbolic system is causal reasoning.
neural_symbolic:
layer_1: neural
- object detection (what is there?)
- pose estimation (how is it positioned?)
- motion tracking (where is it going?)
layer_2: symbolic
- context inference (what situation is this?)
- causal reasoning (why is this happening?)
- SQL queries over temporal event database
- prediction (what will happen next?)
- explanation (human-readable "why")
layer_3: pda # LLM interface layer
- generate: natural language → SQL
- perform: execute queries
- interpret: results → meaning
- explain: causation in plain language
- visualize: charts, timelines, maps
- remember: query history, preferences
The neural layer provides perception. The symbolic layer provides reasoning. The PDA layer provides natural language interface -- neural at the surface, symbolic in the protocol.
Every inference follows Drescher's schema pattern:
schema:
context: [observable conditions]
action: [event that occurred]
result: [observed outcome]
learning:
marginal_attribution:
- which context features predict result?
synthetic_items:
- inferred entities not directly observed
generalization:
- when does this schema apply elsewhere?
Intelligence at the edge, not in the cloud:
edge_architecture:
edgebox:
location: factory floor
latency: <50ms
capabilities: [inference, alerting, logging]
cloud:
purpose: training, aggregation, analytics
latency: acceptable for non-real-time
principle: |
Real-time decisions happen at the edge.
Learning and optimization happen in the cloud.
Data sovereignty stays with the customer.
safety_monitoring:
purpose: Prevent accidents through predictive awareness
examples:
- pedestrian_in_vehicle_zone
- ppe_compliance (hard hats, vests, glasses)
- ergonomic_risk (repetitive motion, lifting posture)
- near_miss_detection (close calls before accidents)
output:
alert: real-time notification
explanation: why this is a safety concern
recommendation: suggested action
audit: logged for compliance
process_optimization:
purpose: Improve efficiency through observation and inference
examples:
- cycle_time_analysis
- bottleneck_detection
- idle_time_measurement
- workflow_optimization
output:
insight: what is happening
causation: why it is happening
recommendation: how to improve
simulation: what-if scenarios
predictive_maintenance:
purpose: Fix equipment before it fails
signals:
visual: vibration patterns, wear indicators, alignment
thermal: heat signatures indicating friction or failure
acoustic: sound patterns indicating mechanical issues
schema:
context: [equipment state, operational history]
action: [detected anomaly]
result: [predicted failure mode]
output:
prediction: what will fail, when
explanation: why we predict this
recommendation: maintenance action
confidence: certainty level
devops:
purpose: Apply MOOLLM patterns to infrastructure
patterns:
files_as_state:
- infrastructure as code
- git as audit trail
- YAML as configuration
coherence_engine:
- detect configuration drift
- propose remediation
- explain changes
speed_of_light:
- batch operations
- parallel deployment
- minimal round-trips
# Factory zone as MOOLLM room
zone:
id: assembly_line_3
type: [production, monitored, indoor]
contains:
- equipment: [robot_arm_1, conveyor_2, station_7]
- personnel: [operator_badge_1234]
- cameras: [cam_3a, cam_3b, cam_3c]
exits:
- to: staging_area
- to: quality_check
atmosphere:
safety_status: green
production_status: active
alert_level: none
# Forklift as MOOLLM character
entity:
id: forklift_07
type: [vehicle, autonomous, tracked]
location: loading_dock_2
state: stationary
current_task: awaiting_clearance
relationships:
operator: badge_5678
cargo: pallet_1234
needs:
fuel: 0.73
maintenance: 0.15 # due soon
# Safety protocol as MOOLLM skill
skill:
id: pedestrian-safety
activation:
context: pedestrian detected in vehicle zone
action:
- alert vehicle operators
- log safety event
- track pedestrian until zone-clear
advertisement:
provides: pedestrian-zone-monitoring
satisfies: [safety, compliance, awareness]
| Team Member | Role | Background |
|---|---|---|
| Henry Minsky | CTO | MIT AI Lab, NTT DoCoMo, Google Nest. Marvin Minsky's son. |
| Dr. Cyrus Shaoul | Chief Evangelist | Computational neuroscientist, Digital Garage co-founder/CTO |
| Dr. Milan Singh Minsky | VP Product | Venture-backed startups, RayVio co-founder |
| Sheung Li | VP Applications | Machine vision in manufacturing |
| Dr. Steve Kommrusch | Senior AI Research Scientist | Deep learning, AMD/HP/National Semiconductor |
| Don Hopkins | AI Architect | The Sims, NeWS, pie menus, MOOLLM |
The theory meets the practice. Minsky's ideas, refined through Hopkins's implementation experience and Kommrusch's deep learning expertise, deployed on factory floors.
transparency:
principle: Every inference is explainable
implementation:
- causal_chains: visible in audit log
- confidence_levels: always reported
- uncertainty: acknowledged, not hidden
- limitations: documented
privacy:
principle: Data sovereignty and minimal collection
implementation:
- edge_processing: data stays local when possible
- anonymization: faces pixelated by default
- retention: minimal, configurable
- consent: clear signage, worker awareness
human_agency:
principle: AI advises, humans decide
implementation:
- critical_decisions: require human approval
- recommendations: clearly labeled as suggestions
- override: always possible
- accountability: human remains responsible
| System | Integration |
|---|---|
| SCADA | Sensor data ingestion |
| MES | Production event correlation |
| ERP | Business context enrichment |
| CMMS | Maintenance recommendation routing |
| Safety Systems | Alert escalation |
deployment:
edge:
edgeboxes: industrial compute at the source
latency: <50ms for real-time inference
resilience: operates offline if cloud disconnected
cloud:
platform: customer choice (AWS, GCP, Azure, on-prem)
purpose: training, aggregation, dashboard
sovereignty: customer owns their data
hybrid:
edge_to_cloud: telemetry, events, learning data
cloud_to_edge: model updates, configuration