arcana.studio中文

Do more before you hire more.

Senior AI builders, directly involved from workflow design to production.

Let’s talk
Recently delivered: a batch short-drama pipeline →
01 / Selected work

Real needs. Reliable delivery.

Content production · Delivered

One long drama. A batch of shorts.

The client sends the episodes and subtitles. The pipeline reads the story frame by frame, splits scenes, plans each cut, then adds bilingual subtitles and music. A person reviews every clip before delivery.

61 → 28
episodes across 3 series, cut into 28 vertical clips
2–3 min
per clip, cut along story beats
ZH / EN
bilingual subtitles, music planned per segment
3 days
to deliver the first batch
PalaceThe court entry cut
RomanceThe KTV test cut
FamilyThe breaking-point cut
PalaceThe snake-revenge cut
RomanceThe fake-boyfriend cut
RomanceThe walk-away cut

Excerpts from delivered clips, muted and looping. Each label is the plan name the pipeline gave that cut.

Every step leaves something you can check

  1. 01Read every frame

    One frame per second, described in plain language.

    Ep 4 · 00:01Five men in an ornate private room around a white table set with beer and flowers.
    Ep 4 · 00:05A young man in a black Chinese-style jacket, arms crossed, looking straight at the camera.
  2. 02Split scenes

    Subtitles and frames together turn each episode into story scenes.

    Scene 02 · She arrives lateEp 4 · 00:23–00:44

    She walks into the room; the camera cuts between her entrance, the others' reactions and the seating.

  3. 03Group the story

    Adjacent episodes become one complete arc with a set amount of footage.

    Group 02 · Power shiftEps 4–6 · ~557 s

    Private pressure in the first half, a public workplace takeover in the second, split at the appointment.

  4. 04Plan the cut

    Two cuts per arc, each with its hook, length and trade-offs written down.

    The KTV test cutTarget 150–210 s

    Open on the drinking-game dare and compress the setup; keep the hallway warning; end on her talking herself down.

  5. 05Finish and review

    Cut to plan, add bilingual subtitles and segment music, then review each clip by hand.

    Format
    1080 × 1920 vertical
    Subtitles
    Chinese + English
    Music
    Planned per segment
    Per arc
    2 versions

Series footage belongs to its rights holders. This page shows excerpts of pipeline output only.

AI operations inspection · Demo

Routine checks. An agent that follows through.

An agent operates your existing dashboard, checks device status, and records the evidence behind each finding.

Defined workflowsException checksEvidence captured

Illustrated workflow based on the project. Device names are simplified; people handle follow-up investigation.

Workflow demonstrationAgent signing in
Inspection AgentAgent workflow · Demo

Task: check device status and record exceptions.

Action: open the business dashboard

Business dashboard / Devices

Device status check

Review device status against agreed criteria.

DeviceStatusEvidence
Device ANot queried—
Device BNot queried—
Device CNot queried—

The agent starts a task in the existing dashboard.

00 / 24 s

01 / Agent opens the dashboard

02 / Our approach

Behind every delivery, a reliable method and system.

Usable outcomes. A process you can count on.

  1. 01

    Scope & standards

    Goals / scope / acceptance criteria

  2. 02

    Prototype & validate

    Prove the key outcomes

  3. 03

    Create & develop

    Build a repeatable process

  4. 04

    Test & review

    Automated tests + human review

  5. 05

    Deliver & iterate

    Handover / support / improvement

Refine based on test results

Clear standards

Agree on outcomes, scope, and deliverables up front.

Verifiable quality

Check content and system performance against real scenarios.

Continuous improvement

Support evolving needs and long-term operation.

03 / Working together

Choose the way we work together.

04 / Team

Meet the people who build it.

Engineering experience in deep learning and big data systems, applied to real delivery.

Portrait of Baoshiqiu

Baoshiqiu

Expert Engineer

I deliver end-to-end deep learning and big data systems.

Kuaishou
Real-time deep learning systems for video processing and real-time visual effects rendering.
JD.com
Large-scale processing of search and recommendation data using Flink.
DiDi
Real-time perception systems for autonomous driving and deep learning inference optimization.