WaveCraft

Ι — Origin

How it started.

WaveCraft began as a college project in May 2026. But the underlying problem felt much larger. Creating content is difficult — not because ideas are difficult, but because perfectionism becomes a barrier. People hesitate to begin. People hesitate to publish. Professional software can feel intimidating. Workflows can feel endless. Learning curves can feel exhausting. WaveCraft emerged from a simple question: what if editing felt conversational?

ΙΙ — The Problem

What we're solving.

Recording content has become easy. Editing remains difficult. People spend hours cleaning recordings, learning workflows, exploring plugins, watching tutorials, managing exports and versions. Many creators stop before publishing — not because they lack creativity, but because tools create hesitation.

People don't want compressors. People don't want EQ curves. People don't want plugins. People want outcomes. Make this podcast sound professional. Remove noise. Fix mistakes. The workflow should emerge from intent — not the other way around.

ΙΙΙ — Differentiation

Why this, not that.

Unlike generic AI tools or traditional DAW software, WaveCraft bridges the gap between intent and execution. We use a deterministic feedback loop — if the AI makes a mistake, it catches its own clipping errors via mathematical DSP audits, not "vibes". The pipeline is transparent: WhisperX for transcription and alignment, pyannote.audio for speaker diarization, SAM for source separation — each step auditable, each output inspectable. A FastAPI layer handles state-free routing while a Ray GPU cluster serializes heavy model work, preventing the OOM crashes that plague naive multi-model pipelines.

ΙV — Capabilities

What it does today.

• ›Speaker diarization (pyannote.audio)
• ›Transcription + word-level alignment (WhisperX)
• ›Source separation (SAM)
• ›Spectral noise reduction (SNR-gated)
• ›LUFS mastering to broadcast standard
• ›Intent-driven pipeline orchestration (LangGraph + Gemini)
• ›Branching outputs and workflow graphs
• ›Human-in-the-loop modifications
• ›S3-streamed uploads (no disk locking)
• ›Real-time progress via offset polling (SSE-safe)

V — Target Audience

Who we're building for.

VI — Open Questions

What we don't know yet.

• Do people want an audio editor, or do they want broader media tooling?
• What workflows matter most?
• How much automation feels useful versus intrusive?
• Do users care about conversational editing or branching outputs?

Feedback wanted from: Audio engineers, podcast creators, media professionals, video editors — and people who abandoned content creation.