Smakosh®

LLM features that survive production.

Chat, agents, MCP servers and multi-provider routing built into your product by the engineer behind LLM Gateway, with cost, latency and failure handled from day one.
4K+
paying customers on the gateway I build and run
1.6K+
GitHub stars on the open-source core
5
provider companies onboarded through Airside
SOC 2
Type II compliance program led

01What you get

The outcome, not the hours.

Who it is forProduct teams adding AI features, and platform teams who need one reliable way to reach many model providers.

  1. 01

    An AI feature users can rely on

    Streaming, retries, fallbacks and timeouts designed in, so one provider outage is not your outage.

  2. 02

    Costs you can predict

    Usage tracked per customer and per feature, with caching, budgets and the right model for each job.

  3. 03

    No lock-in to one model vendor

    One OpenAI-compatible interface in front of every provider, so switching is a config change.

  4. 04

    Something your team can maintain

    Typed code, evaluations for the prompts that matter, and documentation for how it all fits.

02Pricing

Fixed prices, written scope.

No hourly meter and no surprise invoice. You see the price before the work starts, and it does not move once it has.

  1. 011 week

    LLM architecture review

    A second opinion on your AI feature or platform before it gets expensive.

    • Review of prompts, routing and data flow
    • Cost, latency and failure analysis
    • Provider and model recommendations
    • Written report and a 60 minute readout

    $2,400

    fixed price

    Book this package
  2. 022 weeks, renewable

    AI feature sprint

    Where most projects start

    One AI feature built into your product and shipped: chat, an agent, search or an MCP server.

    • End to end in your stack, in your repo
    • Streaming UI, tool calls and structured output
    • Evaluations and usage tracking included
    • 14 days of follow-up after the sprint

    $9,000

    per 2-week sprint

    Book this package
  3. 034 to 8 weeks

    Gateway and platform build

    Multi-provider routing, keys, billing and analytics as an internal platform or product.

    • Provider routing with fallback and caching
    • API keys, budgets and usage analytics
    • Billing integration where needed
    • Self-hosted or built on LLM Gateway

    from$18,000

    fixed price

    Book this package

Prices in USD, excluding VAT where it applies. Model usage is billed to your own provider accounts. The review price comes off a sprint booked within 30 days.

03How it runs

Four steps, no mystery.

  1. Step 01

    Architecture review

    What you want the feature to do, the data it needs, and the failure modes worth designing for.

  2. Step 02

    Thin slice

    The smallest end-to-end version in production behind a flag within the first week.

  3. Step 03

    Harden

    Evaluations, rate limits, cost tracking, fallbacks and the unhappy paths.

  4. Step 04

    Handover

    Docs, dashboards and a walkthrough so your team owns it after I leave.

04Background

The long version.

Getting a demo to answer a question takes an afternoon. Getting a feature to answer ten thousand of them a day, at a cost you planned for, when a provider is having a bad hour, is the actual job.

That second part is what I do all day.

#Where this comes from

I co-founded and build LLM Gateway, an open-source AI gateway: one OpenAI-compatible API in front of many model providers, with routing, fallback, caching, usage analytics and billing. I work across the gateway API, the account and billing backend, the customer dashboard, the docs, the chat playground and the provider portal, in TypeScript, Next.js, Hono, PostgreSQL and Redis.

Around it I have shipped:

  • DevPass, flat-rate model access for coding tools, with subscription management, usage dashboards and pay-as-you-go overflow.
  • Airside, a self-service portal where model providers onboard themselves: domain verification, capability checks, pricing approvals, routing controls and traffic analytics.
  • DevPass Code, an open-source terminal coding agent with browser authentication and unified model access.
  • @llmgateway/ai-sdk-provider, so teams on the Vercel AI SDK can adopt it in one line.

I also led the SOC 2 Type II program for it, which shapes how I think about keys, access and data on every client build.

#What you typically hire me for

  • Adding a chat, search or agent feature to an existing product without rewriting the product.
  • Building an MCP server so assistants can use your product safely.
  • Replacing a tangle of per-provider SDK calls with one routed, observable interface.
  • Bringing an AI bill under control with caching, model selection and per-customer limits.
  • Designing usage-based billing for an AI feature, so pricing follows cost.
  • A hard look at an AI roadmap before you commit two quarters to it.

If you are about to build on top of a model, or already have and it is starting to hurt, send me the details.

05Questions

Asked before signing.

Which models and providers do you work with?

OpenAI, Anthropic, Google, and the wider set I route to daily through LLM Gateway. I pick per task on quality, latency and price, and keep the choice swappable.

Do we have to use LLM Gateway?

No. I will recommend it where it fits and say so when a direct integration or another tool is the better call.

Can you build an MCP server for our product?

Yes. That is a common sprint: a typed MCP server that exposes your product's actions and data to AI assistants, with auth and rate limits.

How do you handle sensitive data?

We decide up front what may leave your infrastructure and to which provider. Where it cannot, I design for self-hosted or zero-retention options.

What does it cost to run afterwards?

You get a cost model per feature before launch and usage tracking after, so the bill is a number you expected.

Tell me what you are building

A first call is 30 minutes and free. You leave with a view on scope and price whether or not we work together.

Book a scope call