Connect the operational picture
Read-only integrations with BMS, CRM and LabMall bring order, procurement, warehouse and follow-up data into one query workflow.
LD Assist
One conversation. A connected view of fulfillment.
An enterprise fulfillment agent powered by a locally deployed large language model. Connect orders, purchasing, warehouse records and follow-ups so support teams can check progress, confirm batches and prepare reply drafts.
Independently designed, built and delivered an enterprise fulfillment agent—from requirements and product design to full-stack development, testing and launch.
- support users
- 30
- queries per day
- ~300
A better way to work
BEFORE
Support teams switch between systems to identify an order, match its products and purchase batches, and piece together warehouse and follow-up records.
WITH THE PRODUCT
A single conversation guides users through product and batch selection, retrieves the relevant records and keeps follow-up questions in context.
Core capabilities
Resolve the right product and batch
Explicit choices help users disambiguate multiple products or purchase batches before the assistant proceeds.
Keep the conversation grounded
Follow-up queries reuse the confirmed business objects and retrieve their current records, covering quantities, shipping progress and sources.
Turn findings into a useful reply
Visible progress, linked sources and editable reply drafts help support teams act on the result. History recovery and exception messages support everyday use.
Local model deployment and integration
Deploy a large language model on purchased GPU hardware and connect local inference to LD Assist, linking business queries, agent orchestration and reply generation from the model runtime through to the application.
How it works
- 01
Ask a fulfillment question
- 02
Confirm product and batch
- 03
Query connected systems
- 04
Review and follow up
Key design decisions
Confirm the object before querying
An order can include several products and purchase batches. I put explicit selection before retrieval so support staff choose the right object before warehouse and follow-up records are connected. Further questions stay tied to that selection.
Make replies traceable to their sources
Support staff need fulfillment details they can check. Read-only queries, visible sources and reply drafts work together so staff can review the evidence before deciding how to respond to a customer.
Keep context; retrieve records again
Follow-up questions reuse the confirmed order and batch without making staff select them again. The workflow retrieves the relevant records afresh, keeping the conversation continuous and quantities and shipping updates grounded in business data.
PRODUCT / ENGINEERING
What I delivered
Define the problem
Map support workflows, product scope, business objects and query paths.
Design the product
Design conversation flows, candidate selection, follow-up questions and exception handling.
Full-stack development and model deployment
Build the mobile web frontend, backend, agent orchestration, data integrations and state recovery. Deploy the model on purchased GPU hardware and integrate local inference into the application.
Test and launch
Establish unit tests, integration checks, CI and the release workflow.
I use LangGraph to orchestrate queries, with permission checks and deterministic business rules for dates, quantities and critical choices. Conversation state recovery, unit tests, integration checks and CI support ongoing iteration and release.
TRY THE WORKFLOW
INTERACTIVE WALKTHROUGHWalk through a complete query.
This walkthrough uses sample data and preset results, without connecting to company systems or live AI.
- 01 Ask
- 02 Select product
- 03 Select batch
- 04 Review & follow up
Start with a question from everyday work
Choose the sample question to explore order lookup, confirmation and a draft response.
ORDER FOUND · DEMO-2026-001
This order has two products. Which one do you need?
Confirm the item so information from different products stays separate.
Two purchase batches found. Choose one.
Batches may be at different stages. The result will reflect the batch you select.
LOOKUP COMPLETE · PRESET EXAMPLE
Review the facts. Then draft a reply.
- Product
- Batch
- Status
- Tracking
ASK A FOLLOW-UP
Does your team face a similar challenge?
Start with one workflow. Let’s explore how AI could fit your operations.
Keep exploring
All work
跟一下
Personal follow-up app0→1 · Solo build
跟一下 (Genyixia)
A lightweight iOS app for replies, refunds and documents that still need follow-up. I independently designed and built the product, connected its paid features and shipped it on the App Store.
- Available now
- App Store
- Native app
- iOS
- Subscription & lifetime
- Pro
Intelligent warehouse label capture0→1 · Solo build
WMS Label Recognition
Extract manufacturer lot numbers, manufacture dates and expiry dates, with a standard API connecting recognition, human review and WMS entry.
- recognition accuracy
- 100%
- logistics users
- 28
- calls per day
- ~520