How Baidu’s One‑Stop Search Platform Cuts Development Costs by 80%

Background

Baidu’s vertical‑search architecture team supports hundreds of search scenarios across dozens of business lines. Over years of growth they accumulated many developer platforms and tools covering retrieval, presentation, strategy, and offline modules.

Problems

High maintenance cost : fragmented R&D platforms lack unified engineering infrastructure and model abstraction, causing duplicated effort and costly legacy stacks.

Steep integration learning curve : a long, complex search system chain with many modules makes onboarding and development cumbersome.

Goals and Approach

The team aimed to (1) build more efficient engineering infrastructure to lower architecture‑side R&D and operations costs, and (2) abstract and model the business‑side search development to reduce usage and iteration barriers, delivering an end‑to‑end one‑stop service.

Solution Overview

FaaS mechanism : developers focus on business logic while the platform provides online debugging, deployment, capacity management, and ops control, shifting from service development to function development.

SaaS services : standardized search‑service packages are built from reusable components, enabling low‑code, visual integration for business teams.

One‑Stop Application Center

The platform unifies existing tools into a single entry point. Role‑based navigation shows relevant capabilities for developers or operators, and users can customise favourites.

FaaS Mechanism

The team selected Baidu Cloud Function Compute (CFC) as the underlying FaaS platform and extended it with business‑oriented management capabilities.

Performance Optimisations

Custom Go SDK : built on the Go Develop Platform (GDP) to cache Baidu Naming Service (BNS) results in memory and provide a BNS acceleration service, reducing lookup latency.

Function pre‑warming : keeps at least one container alive to eliminate cold‑start delays for hot functions.

API Gateway and Capability Extension

The platform offers visual configuration of HTTP‑to‑function routing, optional authentication middleware, and the ability to integrate external services via custom server deployments when FaaS cannot meet complex requirements.

SaaS Service Productisation

Standardised search‑service packages (e.g., text relevance, semantic vectors, structured search, vector DB) are built as composable building blocks. A JSON‑Schema‑driven dynamic form system (using FormRender and Ant‑Design) enables self‑service package publishing.

Custom Widgets

When standard FormRender components are insufficient, custom React widgets are created to handle async loading, file uploads, and inter‑field dependencies.

DAG Visualisation

Both search‑service packages and presentation‑layer pipelines are modelled as Directed Acyclic Graphs (DAG). Reusable DAG visualisation components support drag‑and‑drop editing, improving understandability and iteration speed.

Summary and Outlook

By addressing high maintenance cost and steep integration barriers through a unified FaaS‑SaaS platform, Baidu achieved an 80% reduction in development cost, 95% savings in server/bandwidth operations, and more than 4 million calls per month across 300+ functions. The solution now serves over 180 daily search developers, demonstrating a repeatable, open‑ecosystem approach for accelerating complex business scenarios.