A Bistronomy Brand,
Built from Scratch.
Haiba Bistronomy is a seasonal, demand-driven pop-up restaurant brand for premium Southeast Asian fusion, conceived, designed, and launched by Fauzul. From generative AI food prototyping to a serverless ordering platform, every detail of this brand was built by one person.
Southeast Asian fusion, done right.
Dhaka's upscale dining scene is saturated with repetitive pan-Asian, Japanese, and Thai concepts. What was missing: authentic, premium Indonesian-Malaysian bistronomy — the kind of bold, spice-forward food paired with the sleek presentation of a European fine-dining experience.
Fauzul saw this gap and built Haiba from zero — launching in November 2025 with 4 successful pop-up runs to validate real demand before committing capital. Today he runs it as a seasonal, demand-driven pop-up.
Every dish was prototyped in AI before it hit the kitchen.
Traditional restaurant launches burn thousands on physical kitchen R&D and commercial photographers. Fauzul skipped both by building a visual prototyping pipeline using custom Nanobanana and Seedream generative image models — generating hyper-realistic plate layouts as direct blueprints for the chef.
The recipes themselves were designed by Fauzul through AI-assisted research, then personally taught to the head chef and refined to perfection using local Bangladeshi ingredients.
Vietnamese Rolls — AI-Generated Target
Before lighting a single burner, every signature dish was prototyped as a pixel-perfect generative image. These AI outputs — like the Vietnamese Rolls with Peanut Sauce above — became the exact visual target the chef aimed to recreate, replacing expensive commercial food photography entirely.
From AI Target to Real Plate
The generative image was the target — the flavor was engineered separately. Fauzul designed precise recipes through AI-assisted research and deep culinary study, then personally taught them to the head chef. Each dish was iterated and refined using local Bangladeshi ingredients until it matched the visual prototype perfectly — with zero physical R&D food waste.
Straight to Social, Zero Photography Budget
The generative outputs became the actual marketing assets. Dishes like the 12" Oceanic Pizza and Sambal Wings were published directly to social media and the ordering catalog — establishing a premium, consistent brand identity from day one and saving over $10,000 in commercial photography and videography.
4 Pop-Ups → A Seasonal Pop-Up
Starting November 2025, Haiba ran 4 successful pop-up events to validate demand and crowd acceptance before committing to a lease. It now runs as a seasonal, demand-driven pop-up.
Ordering, state management, and serverless infrastructure.
The ordering experience is built on local-first state management (Zustand + localStorage). When a guest places an order, the UI transitions immediately — giving instant visual confirmation and live status updates — while the Go backend asynchronously completes the transaction via Google Cloud Run.
This means no mandatory sign-ups, no invasive tracking. The local state remembers past preferences and tracks orders using a predictive delivery timeline, so repeat diners can re-order instantly without ever creating an account.
On the infrastructure side, the frontend is statically hosted on Cloudflare Pages at the edge. The backend runs a Go + ConnectRPC microservice on Google Cloud Run, which scales to zero during off-peak hours — so compute cost perfectly mirrors actual restaurant operating hours.
Real numbers, real community.
If we worked together.
I bring high-leverage product engineering and absolute operational discipline: shipping weekly, pressure-testing user inputs, and automating serverless cloud deployment pipelines.
See it in the wild.
This case study covers how Haiba Bistronomy was built — these links open what ships today: the live product, repo, demos, and supporting material.
The changelog.
VisaPros
Submitted as the capstone for the Google × Kaggle 5-Day AI Agents Intensive (2025). VisaPros is an agentic visa eligibility advisor on a Google ADK agent-to-agent (A2A) hub-and-spoke architecture: it parses documents once, then runs six country agents in parallel, each its own A2A server, to evaluate fit across destinations.