video strategy extraction

The problem

Instructional videos and expert walkthroughs are easy to record but difficult to use as technical specs. Manual transcription and extraction lead to "hallucinated" parameters where an AI fills in gaps with plausible but unstated values.

What it does

• Transcribes audio using local Whisper or Deepgram with a focus on propositional content.
• Extracts structured rules and parameters into a schema-validated JSON spec.
• Validates every extraction against a strict nullability contract to prevent fabricated data.
• Calculates confidence scores based on technical specificity rather than speaker fluency.
• Filters out motivational noise, non-speech, and opinion-only content with explicit status records.

Frameworks & tools

Faster-whisper, Deepgram, Groq, Pydantic, and any LLM capable of structured output (Hermes-class or better).

Why this beats prompting it yourself

Generic prompts "smooth" over missing information, creating confident-sounding specs that didn't exist in the source. This skill enforces Iron Rules that prioritize precise gaps over fluent fabrications, ensuring you never build a system on an assumed parameter.

Use cases

• Convert trading strategy videos into backtestable entry/exit rule sets.
• Turn recorded software onboarding walkthroughs into actionable SOPs.
• Transform training videos into machine-checkable compliance checklists.
• Process folder batches of expert demos into structured technical runbooks.

Known limitations

Requires an instructional or procedural source. It will explicitly refuse to generate specs for motivational or opinion-heavy content without extractable steps.