Skip to content
Shenzhen · The Greater Bay Area · Earth

B2B Agency

Architecting AI-Native B2B Workflows.

Delivering end-to-end design engineering and autonomous systems for global enterprises.

Book a Strategy CallFrom $200 — a scoped de-risking sprint before any commitment

Capabilities

Three pillars of delivery.

  • SLOT A

    Autonomous AI Workflows

    Integrating LLMs into your existing business logic. From multi-agent orchestration to custom MCP server deployment.

    Deliverables

    • Architecture brief: where the model belongs, where it must not be trusted
    • Custom MCP servers exposing your internal tools to any compliant client
    • Multi-agent pipelines with typed tool schemas and deterministic fallbacks
    • Evaluation harness with a regression set before anything reaches production
    • Cost model per workflow and per seat, with hard ceilings and kill switches
    • Observability: traces, token accounting and failure replay
  • SLOT B

    Design Engineering

    Bridging the gap between Figma and production. Delivering pixel-perfect Next.js/React front-ends with robust data architectures.

    Deliverables

    • Production Next.js App Router implementation, statically rendered where possible
    • Design tokens as typed code, with a single source of truth for both themes
    • Data layer: schema design, caching strategy, and loading/empty/error states that were designed rather than discovered
    • Core Web Vitals budget enforced in CI, not aspirational
    • Accessibility to WCAG 2.2 AA on the flows that matter
    • A component library your team can extend without asking me
  • SLOT C

    UI/UX Systems Architecture

    Moving beyond isolated screens. Crafting scalable design systems and Liquid Glass aesthetic interfaces for complex B2B SaaS products.

    Deliverables

    • Information architecture rebuilt around the operator's actual decisions
    • Density strategy: what is always visible, what is one interaction away, what never ships
    • Interaction model for asynchronous AI work — progress, partial results, interruption, undo
    • Role-based surface design for multi-tenant products
    • Governance: naming, versioning and deprecation rules for the system itself
    • Written decisions log, so the reasoning survives staff turnover

Featured Work

Systems, not screens.

  • Chefshot — own product

    An outbound engine that diagnoses a restaurant before contacting it

    The Challenge

    Chefshot makes commercial food photography without a studio, and the product works. The problem was that a restaurant owner does not wake up wanting better photography — they want more orders. Their menu photos are, in their mind, fine. The gap between what they have and what changes their conversion rate is invisible to them, and no campaign closes an invisible gap. Doing the diagnosis by hand was not viable either: judging a restaurant's photography properly takes a minute or two of real attention, and writing a specific observation takes several more, which is five minutes per prospect for a message with a single-digit reply rate.

    Stack

    • n8n
    • Dify
    • Apify
    • Apollo.io
    • Twenty CRM
    • PostgreSQL
    • Vercel AI Gateway

    Engagement: Own product: architecture and engineering by me, ongoing.

    Read the full architecture and where it breaks at 100k

    The Architecture

    • A search matrix — city × cuisine × qualifier — built in code at the head of the pipeline, so the targeting decision is versioned and reviewable rather than scattered across a scraper config
    • Batched scraping, one matrix combination per run, so a single query never has to survive a timeout
    • A deduplication funnel that checks each venue against the CRM by stable Place ID before any paid API touches it — the cost control that makes the economics work
    • An inverted image-triage step: a fast vision model is asked to choose the WORST dish photo from a candidate pool, because a restaurant's worst hero-dish shot is the one costing them orders
    • Diagnosis through a Dify agent with a retrieval knowledge base holding the photography standard and a library of before-and-after cases, so the critique cites a rule rather than an opinion
    • A parallel enrichment branch (company fundamentals and decision-maker contacts) that writes Company → Person → Diagnosis into the CRM in that order, so no record is orphaned

    The Impact

    • One person operates the pipeline: the diagnostic judgement, which is the only genuinely expert step, runs at machine speed while everything downstream stays reviewable
    • Outreach carries a specific, checkable critique of the recipient's own photograph, which is what distinguishes it from a campaign
    • Deduplication before any paid call keeps unit cost proportional to genuinely new prospects rather than to scrape volume
    • The design has a known ceiling, calculated rather than assumed: about 300 venues an hour, so ten thousand takes roughly a day and a half, and a hundred thousand would take two weeks of continuous execution — at which point the orchestration has to be replaced with a queue and a stateless worker fleet rather than tuned

Engagement model

Engagement model

Fixed-scope phases with a written architecture brief before any code. Pricing is quoted per system, not per hour, so scope creep is a shared problem rather than a hidden invoice.

  1. 01

    Architecture brief

    A short written document before any code: the bottleneck, the decision that unlocks it, the shape of the system, and the order-of-magnitude cost. It is yours even if we stop here.

  2. 02

    Fixed-scope build phase

    Two to ten weeks, one written scope, weekly working software. No phase ends without something running in a real environment against real data.

  3. 03

    Handover and operating loop

    A decisions log, the runbooks, and a recorded walkthrough. Then a retainer only if the system genuinely needs ongoing tuning — not by default.

What this is not

  • Hourly body-shopping. I quote per system because an hourly rate makes the interesting decisions financially irrational for both of us.
  • Staff augmentation inside a process I cannot influence. If the architecture is already fixed and I am only permitted to implement it, a contract developer is the cheaper and more honest hire.
  • Screens without a system. If the deliverable is a set of mockups that nobody has agreed how to build, the project will fail at handoff regardless of how good the mockups are.
  • Projects where the honest answer is "you do not need AI yet". I will say so on the first call instead of selling a model into a process that needs a data model first.

Secured & escrowed via Upwork Direct Contracts.

Shenzhen · Working languages: EN / 中文

Initialize Your System.

A short technical call is enough to establish whether this is a fit.

Book a Strategy Call

Ready to build a system?[ Book a Call ]