SEEDESTINY / APPLIED AI STUDIO
Take AI out of the demo,
into your business processes.
We design and deliver knowledge systems, automated workflows and custom AI applications for cross-border trade, content teams and professional services. Every delivery includes source code, documentation and the ability to keep iterating.
Validate on a small scope first, then decide whether to scale up.
PRINCIPLES
How we work
-
01
Real tasks before big platforms
Start from an email, a contract, a quotation or a content production line. Prove it works on one real task first, then decide whether to expand.
-
02
People keep the final say
Retrieval comes with evidence; automation has boundaries. Outbound email, key judgments and high-risk actions keep human approval and a rollback path.
-
03
No black-box handoffs
Source code, configuration, usage docs and validation records are delivered together. You can modify things yourself — or have us keep iterating.
SELECTED WORK
Project evidence, not concept demos
Every project is labeled with its true status: running, live, internal, prototype or iterating. Numbers cover only what can be verified.
Grid Office: data-driven AI animation production
Connecting real A-share closing data, character assets, voice and video generation into an original animation series that keeps updating.
- 60+ episodes and still updating
- Driven by real trading-day data
- Reusable character & voice assets
Ozon compliance knowledge base
Platform contracts, official courses, revision records and the Russian parallel-import list organized into a compliance resource system you can search, trace and set deadline reminders in.
- 38 official courses
- 468 list entries
- 2,321 brands
Pathfinder rule engine & character builder
Complex tabletop RPG rules turned into an app that can query, validate and export character sheets, grounding natural-language concepts in sourced builds.
- 15,871 base entries
- 1,318 class variants
- Level 1–20 paths for 55 classes
Email & quotation approval automation
Inbound archiving, material organization, quotation documents and outbound approval connected into one flow — with a hard rule that nothing sends without approval.
- Scheduled inbound archiving
- Templated quotation documents
- Human approval before sending
MOJO Word Garden
A vocabulary app for children's early English learning — word lists grow around each textbook level, and the learning experience is a small world kids actually want to enter.
- Live and publicly accessible
- Word lists expand by textbook level
- Child-friendly learning interactions
Card sticker translator
A Chinese–English glossary built from the rulebook, OCR localization and translation of card scans, and printable A4 sticker-strip PDFs as output.
- Rulebook-based glossary approach
- OCR localization pipeline
- A4 print output target
SERVICES
Four deliverable capabilities
No vague 'empowerment': each capability comes with concrete deliverables and projects that have already been validated.
-
S-01
AI workflow automation
Keep repetitive flows moving, with you still approving the critical actions.
- Email triage & archiving
- Quotation generation
- Scheduled tasks
- Human approval
-
S-02
Enterprise knowledge & compliance systems
Answers that trace back to the source; changes that someone keeps track of.
- Document parsing
- Clause risk index
- Cited evidence
- Version tracking
-
S-03
Custom AI apps & rule engines
From a one-line requirement to a tool you can actually operate.
- Rule engines
- Natural-language grounding
- Character / document generation
- OCR localization
-
S-04
Content & data production pipelines
Connect data, voice and visuals into a repeatable production flow.
- Data-driven scripts
- Character asset libraries
- TTS
- Video generation
PROJECT INDEX
All projects at a glance
| No. | Project | Capability | Status |
|---|---|---|---|
| P-01 | Grid Office: data-driven AI animation production | Content & data pipelines | Ongoing |
| P-02 | Ozon compliance knowledge base | Enterprise knowledge & compliance | Actively maintained |
| P-03 | Pathfinder rule engine & character builder | Custom AI apps & rule engines | Usable build |
| P-04 | Email & quotation approval automation | AI workflow automation | Internal |
| P-05 | MOJO Word Garden | Custom AI apps | Live |
| P-06 | Card sticker translator | Custom AI apps & OCR | Prototype |
| P-07 | Multi-agent collaboration workspace | AI infrastructure & private deployment | Internal |
| P-08 | Logistics quotation & client collaboration workbench | AI workflow automation & data apps |
APPROACH
From a real sample to a capability handover
We don't start with 'platform selection'; we start from one real task. Every step has clear outputs and boundaries.
-
01
Take one real sample
Pick a frequent, time-consuming, error-prone task; bring a sanitized sample and your current process. First define what 'done' means.
-
02
Build a minimal viable validation
Map the data, rules and boundaries, and build a small runnable version. If validation fails, you learn early why — and what the alternatives are.
-
03
Integrate the flow, keep the brakes
Connect to your existing mailbox, spreadsheets, databases or internal systems; outbound sending, high-risk judgments and exceptions keep human approval and rollback.
-
04
Deliver and hand over the capability
We deliver source code, configuration, docs and validation records. You can maintain it yourself — or have us keep iterating in phases.
FAQ
Clear terms before we start
Do we have to build a big platform from scratch?
No. In most cases, starting from one real workflow and your existing tools is safer. Validate the value first, then decide the scope of the system.
Can we avoid being locked to one particular large model?
Yes. Models are chosen by data, cost, deployment and results; for privately deployable cases we assess hardware and maintenance costs separately — and not every requirement suits on-premise deployment.
We already use email, Excel or internal systems — can you still integrate?
Yes. We first confirm whether your systems offer usable APIs, file formats or database paths. Without interfaces, we evaluate semi-automatic flows rather than forcing full automation.
Who owns the source code and documentation?
The delivery scope is written down before we start. Our standard delivery includes the project source code, configuration notes and usage docs — no black-box 'vendor-only modifications' arrangements.
How do we start?
Send an email describing what you repeatedly do, which tools each step uses and where it breaks most often, with one sanitized sample attached. We start by discussing one verifiable small task.
START HERE
Bring the task that costs you the most time or breaks most often — we'll make it small and make it work first.
Chagall Chen · trade@seedestiny.com.cn · +86 18672371212
If AI isn't the right fit, we'll say so directly.