BlueCL / Technology brief

Claude in the product.
Engineering across the stack.

Our use of Claude spans service API workflows, parallel LLM processing, and software development. We connect that experience to MCP tools and the AI infrastructure we are building next.

00. The wider BlueCL portfolio

Mindflow is one concrete Claude API workflow inside a wider portfolio. Across the products below, we use Claude and reusable Skills to move from problem definition to implementation, debugging, documentation, and workflow design. Runtime provider use is labeled separately from development use.

Knowledge workflow

Mindflow MCP

Turn notes into explanations, tasks and structured work connected to Notion.

Claude role: runtime API summarization and extraction

Content / Web platform

Cute Emoji

Make multilingual symbol discovery and copying easier through a focused web service.

Claude role: product iteration, UI and content engineering

Game data / Dashboard

POE2 Exchange

Turn public Path of Exile 2 economy data into searchable prices and history.

Claude role: data workflow, dashboard and feature development

Mobility / Public data

KORAIL Easy

Make train and platform guidance easier to read for a specific travel route.

Claude role: API, UX and implementation assistance

Generative web prototype

Web Vending Machine

Generate a storefront’s content and visual direction from an industry and brief.

Claude role: generative workflow prototyping and engineering

Game tooling

POE2 Overlay

Explore tools that place useful game information near the player’s workflow.

Claude role: rapid prototyping and tool design

Safety / Sensor collaboration

Liflow

Explore AI and sensor signals for unusual situations affecting people living alone.

Claude role: AI workflow design and team development

Education / Finance collaboration

PinGrow

Help young people learn financial products through onboarding and data workflows.

Claude role: backend and AI-assisted development

Claude API runtime implemented Claude used in development Claude integration planned

Shared engineering loop

Problem or user input → service/API or data layer → model-assisted interpretation where appropriate → connected output or developer-facing interface → telemetry, review, and iteration. The exact model and runtime path depend on the product; all products are not represented as live Claude API services.

00. The wider BlueCL portfolio

Knowledge workflow

Mindflow MCP

Turn notes into explanations, tasks and structured work connected to Notion.

Claude role: runtime API summarization and extraction

Content / Web platform

Cute Emoji

Make multilingual symbol discovery and copying easier through a focused web service.

Claude role: product iteration, UI and content engineering

Game data / Dashboard

POE2 Exchange

Turn public Path of Exile 2 economy data into searchable prices and history.

Claude role: data workflow, dashboard and feature development

Mobility / Public data

KORAIL Easy

Make train and platform guidance easier to read for a specific travel route.

Claude role: API, UX and implementation assistance

Generative web prototype

Web Vending Machine

Generate a storefront’s content and visual direction from an industry and brief.

Claude role: generative workflow prototyping and engineering

Game tooling

POE2 Overlay

Explore tools that place useful game information near the player’s workflow.

Claude role: rapid prototyping and tool design

Safety / Sensor collaboration

Liflow

Explore AI and sensor signals for unusual situations affecting people living alone.

Claude role: AI workflow design and team development

Education / Finance collaboration

PinGrow

Help young people learn financial products through onboarding and data workflows.

Claude role: backend and AI-assisted development

Claude API runtime implemented Claude used in development Claude integration planned

Shared engineering loop

01. How we use Claude today

BlueCL uses Claude API in running service workflows and parallel LLM processing. Claude and reusable Claude Skills also support our software development work.

Area	Current use
Service API workflows	Claude API calls support AI features in our services.
Mindflow MCP	Implemented note summarization, task and deadline extraction, MCP tools, and Notion integration.
Parallel processing	Claude summarization runs alongside independent model and parsing tasks, with results combined into a connected workflow.
Development	Claude and reusable Skills support software development and API integration work.

02. Mindflow: from input to a saved result

Mindflow addresses the manual handoff between understanding study notes, identifying tasks, and organizing the result in Notion. Claude supplies the explanatory study content inside a workflow connected to external tools.

Implemented web service / composite path

Text or file inputMindflow web API · connected Notion workspace

Claude APIMarkdown study material · explanations · key takeaways

Structured extraction · parallelA separate model produces title · task · due

Notion outputCreate page → append summary blocks → return page URL

Accept text or a text file.The FastAPI service exposes POST /process and POST /process-file. The implementation uses the signed-in user’s Notion connection and destination database.
Start independent model tasks together.Python workers call Claude’s Messages API for Markdown study material while a separate GPT-based extractor produces title, task, and due JSON. These tasks do not wait for each other to start.
Create the destination, then append the summary.In the implemented fast-save path, extraction results are used to create a Notion page with placeholder content. The Claude summary is converted into Notion blocks and appended when available. User-provided dates and tags can be written as Notion properties; extracted task/date information is also returned by the API. The standard path waits for combined results before saving.
Return structured content and the saved page.The response contains title, task, due date, Markdown content, and Notion status/page information. The user can continue working in the connected workspace.

Claude’s implemented responsibilities

Generate Markdown with headings, explanations, examples, common misunderstandings, and a final key-summary section. A separate Claude extraction tool returns task and due-date JSON. The source uses Claude Haiku 4.5 for these calls.

Connected-tool handling

The Notion writer converts Markdown to blocks, appends them in batches of up to 100, and includes bounded retry/backoff logic for rate-limit and server-error responses. This is implemented handling, not a measured reliability guarantee.

Illustrative input and output

Input: “Review database normalization, including 1NF and 2NF. Submit the comparison assignment by October 16, 2026.”

Claude branch → Markdown study notes: concepts, examples, key takeaways
Parallel extraction → {"title":"Normalization","task":"Compare 1NF and 2NF","due":"2026-10-16"}
Connected output → Notion page with study content + task/date metadata in the API response

An illustrative product scenario, not a captured live response or benchmark. The implemented composite extractor’s prompt requests Korean output. Exact outputs vary with the input and configuration.

These endpoints belong to the Mindflow service implementation; the BlueCL company homepage is a separate site. Workflow code is implemented, with further product development and user validation continuing.

03. Reliability is the next engineering milestone

Our ongoing priorities are Korean-language summary quality, dependable task and deadline extraction, connected-tool reliability, and visibility into API latency and cost.

Quality

Evaluate summaries and extracted tasks against representative inputs.

Orchestration

Improve coordination, error handling, and recovery across model and tool calls.

Cost & latency

Measure workflow timing and API usage to guide practical improvements.

04. AI operations: the problem and target structure

Developers running several inference servers need to understand where requests go, which requests slow down, and how those observations relate to GPU server health. We are developing a real-time LLM load-balancing and GPU-monitoring product around that need.

Request and monitoring plane · in development

The product direction brings request distribution and GPU observations into one dashboard. Its purpose is to help developers inspect their own model-serving environment with operational context.

Claude explanation plane · planned

A planned extension prepares selected metric snapshots and request context for Claude. Claude would explain the observations and propose investigation points alongside the underlying measurements. Provider integration is also planned.

Planned explanation workflow

Requests & telemetryRequest state · processing time · GPU observations

Claude explanation layer · plannedInterpret selected metrics and suggest investigation points

Developer dashboardMetrics alongside explanations · developers decide and act

This is the target design for the Claude extension, not a deployed feature.

Layer	Responsibility	Status
Request distribution	Route LLM requests across inference servers.	In development
GPU monitoring	Collect and present server observations.	In development
Context preparation	Select operational metrics and request context for an explanation.	Planned Claude extension
Claude API	Return understandable explanations and suggested checks.	Planned
Developer control	Display measurements with explanations; developers decide how to respond.	Target design

Claude is intended to support interpretation of operational information. The target extension keeps request routing and infrastructure controls separate from the language-model explanation step.

05. A founder-led AI software company

BlueCL is based in Mokpo, South Korea, and builds on a business established on February 24, 2023. Founder Dongju Kang develops software products and independently conducted research into MCP-based parallel LLM learning support and LLM-based website generation with visual editing.

South Korea business registration number: 494-52-00728

Two 2026 KSCI conference publications document these studies, with publication authors Soojung Lee and Dongju Kang.

Explore our research

Build with BlueCL

Talk to us about AI workflows, connected tools, and LLM infrastructure.

bluecl@bluecl.cloud