Intel stream
Guanlan Think Tank · NeuronSpark
NeuronSpark Media-Tech Limited · Guanlan Agent Library

An agile studio building AI agents and applications.
We start from the problem,
not the technology.

First principles, applied to products: work out why a problem exists before deciding what to build. We have turned that habit into a public developer pain-point board — a continuous read of real user feedback across industries, companies and products, keeping only negative reviews that describe a concrete situation and turning them into needs AI can meet. Companies take them as briefs, hackathon organisers as challenges, and capable developers can sign in to the company intelligence library to see the full evidence and reach the companies involved.

Open the pain-point board →Hackathon challenges
Developers: sign in to the library →
ShippedMEMECMO · Musicraphy · DirectorAI · Live-biologics brand system · CRM for a Vietnamese media group
§1 · DEVELOPER PAIN-POINT BOARD
Real review data · 2026-09-18

Developer Pain-Point & Demand Board

What are users of the leading apps in China, Japan and Southeast Asia complaining about? We continuously collect public negative reviews of leading apps across 13 industries, drop the pure venting, and keep feedback that describes a concrete situation — the kind that points to structural problems an AI agent can fix.

Open the full board →See hackathon challenges
310
Leading products
13,163
Informative reviews
9,855 venting-only reviews removed
42%
Agent-fixable
8
Markets

Industry × pain theme

Support
Task flow
Login & KYC
Billing
Data & sync
Missing features
Stability
Ads
Content
Localisation
Output quality
Privacy
Legal services
Manufacturing
Business services
Media & publishing
Clinics & health
IT products
Brands & retail
Banking & fintech
Consumer apps
AI avatars
Live streaming
Short-video creation
AI assistants
Press any cell to read its numbersColour = share of the industry’s informative reviews in that theme (scaled per row)

Opportunity heat · industry × market

China
Japan
Singapore
Malaysia
Thailand
Indonesia
Vietnam
Philippines
Legal services
Manufacturing
Business services
Media & publishing
Clinics & health
IT products
Brands & retail
Banking & fintech
Consumer apps
AI avatars
Live streaming
Short-video creation
AI assistants
Press any cell to read its numbersColour = sum of opportunity scores in the cell; number = products. Opportunity = pain × agent-fixability × scale × ownership × reachability.

Rising / falling trends need at least two snapshots and will appear here once they exist.

Source: public Apple App Store reviews · updated 2026-09-18 · the share of negative reviews reflects who writes reviews, not overall satisfaction.

§2 · Portfolio

What we have built

Not a portfolio. A library you can call from.

GEO SaaS · flagship product
SHIPPING

MEMECMO

Make your brand visible inside AI answers. Ten specialist agents in one workflow: brand profile extraction → 110 generated buyer questions → cross-model monitoring on five engines (ChatGPT / Gemini / Perplexity / Claude / AI Overview) → the AIGVR five-dimension algorithm (presence 30% · prominence 25% · share of voice 20% · sentiment 15% · citation strength 10%) → optimization guidance → content distribution → encyclopedia strategy.

Roughly 120 real queries × 5 engines per scan. Already serving Southeast Asian DTC brands and companies expanding worldwide. Three subscription tiers.

10 agents · AIGVR · 5 engines
Neural interaction · hardware + software
INTERNAL

Musicraphy

Connects to BrainLink professional EEG hardware and modulates music from live brainwave readings. Electron desktop app with a React UI, personalized music therapy, plus a simulation mode for research and demos.

BrainLink SDK · Electron · React
Content intelligence · automation with taste
EARLY

DirectorAI

Scattered raw recordings → a finished podcast episode. Not mechanical cutting but director-level editorial judgment: reviewing the material, finding the standout lines, planning the narrative arc, cutting the small talk, pacing the emotion.

“Let one person finish what used to take a whole production team.”

faster-whisper · Claude Opus/Haiku · ffmpeg
Strategic composition · cross-domain brand delivery
DELIVERED

Next-generation live biotherapeutics · brand and digital system

Delivered a complete brand site plus three product-line pages for a partner working on gut microbiome testing, FMT 3.0 transplantation, and SafeMatrix™ probiotics. Science writing through visual system through deployment — a full run across an unfamiliar domain, fast.

3 product lines · full brand delivery
GEO · multi-agent orchestration demo
SHIPPING

Guanlan SEA Command Center

An interactive console orchestrating five specialist GEO agents in parallel: T1 corpus scouting, geopolitical compliance audit, regional competitor scanning, six-axis diagnosis, and advanced corpus generation. Live SSE streams each agent’s reasoning as it happens.

5 agents · SSE · terminal-style UI
Strategic composition · enterprise delivery
DELIVERED

A Vietnamese focus-media group · CRM with an AI sales agent

A sales management and customer relationship system designed from scratch, including an AI agent that assists the sales process. Enterprise-grade delivery with localized deployment.

Enterprise CRM · AI sales agent · localized
Cultural technology · a millennium × the present
RESEARCH

Guqin digitization

Turning the musical structure of the traditional guqin into data and visualization. A thousand years of qin scholarship becomes an explorable, interactive digital experience.

Data viz · interactive · cultural tech

The next one could be yours

Got something to build? We compose modules from the library and build it with you.

§3 · Process

Hackathon-style delivery process

From problem to shipping, usually within weeks.

01

Hear the problem

Whatever pain has genuinely gone unsolved, we help you state it precisely

02

Compose from the library

We pick the right modules from four agent families and compose candidate solutions fast

03

Build together

Hackathon cadence, prototyping with you — two to four weeks, not months

04

Ship and evolve

Ship it, iterate on real feedback, and deposit the methodology back into the library

“Every time a run finishes, the methodology is deposited back into the library. The next problem of the same kind starts planning with what we learned last time.”

— DirectorAI’s built-in logic, and how the whole agent library works

§4 · Pathfinders

Pathfinder personas

If you fit any of these four cells, there is probably something we can build together.

Cross-domain founder

Deep in your domain but without an AI team. You want to prototype and validate an idea nobody has tried. Like JunChangYi Bio — explaining microbiome therapy to consumers in the LLM era.

NeedCompose a prototype from the library → ship it fast → optimize while it runs
→ Join the waitlist · Got something to build

Brands that want to be cited inside generative engines

CMOs, growth leads, and the people running overseas expansion. Search is turning into LLM answers, and you want to know how you are described in ChatGPT — and to be described better.

NeedUse MEMECMO SaaS directly · for complex cases we co-build with you
→ Go straight to MEMECMO · app.memecmo.ai

Established businesses embedding AI internally

Traditional media, biotech, cultural institutions. The core business is solid but AI integration is unfamiliar territory, and you would rather not end up with an old system wearing a chatbot skin.

NeedA hackathon team on site, shipping an enterprise agent prototype in two to four weeks
→ Join the waitlist · Enterprise collaboration

Independent creators and researchers experimenting across fields

Music × neuroscience, editing × directorial thinking, traditional culture × digital expression. Too much for one person, and agencies are both expensive and unfamiliar with the material.

NeedA technical partner to hack alongside you, not an outsourcing vendor
→ Join the waitlist · Experimental collaboration
§5 · Library

Agent library · modules we can call on instantly

Not services, not products—capabilities you can compose from.

Neural
Neural Interaction

Brainwave signal capture · biofeedback-driven real-time interaction · hardware, software and content as one

Lead agent
·Musicraphy music-therapy core engine
Composes into
Neurofeedback gamesEmotion–music A/B testingCognitive training appsHardware-driven experimental content
Content
Content Intelligence

Content generation with taste · narrative-structure intelligence · an automated pipeline from raw material to finished piece

Lead agent
·DirectorAI editorial decision engine
Composes into
Video editing agentsBrand narrative generatorsLong-form rewriting and distillationAutomated podcast and interview production
GEO corpus
Generative Engine Optimization

Brand visibility inside LLMs · empirical cross-model measurement · the full chain from diagnosis to content distribution

Lead agent
·MEMECMO ten-agent workflow
·Guanlan SEA six-axis diagnosis
Composes into
Brand AI visibility auditsCross-model competitor comparisonStructural content rewritingRegional GEO rollout
Strategy
Strategic Composition

Working out how the problem should be defined · composing cross-domain modules from the library · assembling agents into client-grade strategic solutions

Lead agent
·JunChangYi Bio brand delivery
·Vietnamese focus-media CRM
·The MEMECMO productization path
Composes into
Cross-domain brand digitizationEnterprise AI-embedded systemsEarly product definition and prototyping
§6 · Founding Genome

Founding genome

Why this agent library looks the way it does.

The architectural genome of Guanlan Think Tank was defined by a single cross-domain synthesis:

Product
Product

Reads client pain quickly Composes modules from the library

Scholar / Writer
Scholar

Eastern cultural training data Narrative logic embedded in the agents’ judgment

Engineer
Engineer

Neural interaction, content intelligence and the actual building of AI agents

Strategist
Strategist

Client-grade strategy Cross-domain composition

All four faculties have been one decision system since day one — not a committee. That is why the agent library grew from the ground up as a heart × craft × field three-layer composable structure rather than four disconnected product lines.

That genome is now being extended by more partners. Joining us, or building alongside us, grows new agent families on the same architecture.

Founder Liu Junshuo · Liu Junshuojeonshoeck.com ↗GitHub ↗
§7 · Join

Find a team that will hack alongside you

Guanlan Think Tank is currently invite-only. The waitlist is reviewed in batches every Wednesday at 09:00 UTC.

Start here

  • Companies: build with usBring a problem nobody has solved properly; get a running prototype in two to four weeks.Join the waitlist →
  • Hackathon organisers: evidence-backed challengesWell-evidenced pain points from the board, packaged with user quotes and judging criteria.See the challenges →
  • Developers: sign in to the librarySee the full evidence and the companies involved, pick a problem, and reach them directly.Sign in →

Already doing GEO? Use MEMECMO directly — the productized flagship GEO SaaS built by Guanlan Think Tank.

app.memecmo.ai ↗

Questions

About NeuronSpark, the pain-point board, and how we work together

What does NeuronSpark do?
NeuronSpark · 观澜智库 (NeuronSpark Media-Tech Limited, registered in Hong Kong) is an agile studio building AI agents and applications. We apply first principles to products: work out why a problem exists before deciding what to build. The Chinese name 观澜智库 is short for 观澜智能体库, the Guanlan Agent Library — every delivery leaves reusable agent modules behind, which is why the next one ships quickly.

Shipped: MEMECMO, Musicraphy, DirectorAI, a live-biologics brand system, and a CRM with a sales agent for a Vietnamese media group.
Where does the developer pain-point board get its data?
From public Apple App Store reviews of leading apps in China, Japan and Southeast Asia. We keep only negative reviews that describe a concrete situation and drop pure venting; every cell and every ranked entry traces back to the original reviews.

One caveat: the share of negative reviews reflects who writes reviews, not overall satisfaction. The board answers “where are people complaining about something specific”, not “is this company good”.
How is the opportunity score calculated?
Opportunity = pain × agent-fixability × scale × ownership × reachability.

Pain is the volume and intensity of informative reviews; agent-fixability is whether an AI agent can actually address the problem — not every pain point qualifies; scale is the size of the product; ownership and reachability decide whether the company could plausibly buy a solution. The ranking in the right column follows this score.
Why are company and product names behind a sign-in?
The public version is anonymised: product names become aliases, and brand names, links and numbers are scrubbed from quotes. Real company and product names, and the potential buyers behind each challenge, are open to paying members and admins only, and are sent only after the server has checked access — a locked browser never receives them; the blur is a visual cue.

That way the problems can be discussed in public without naming turning into public blame.
Where do the hackathon challenges come from, and how can organisers work with you?
A problem becomes a challenge only when it ranks among the main complaints of at least three leading products.

Organisers can ask us to pick well-evidenced pain points from the board to match their theme and tracks, packaged as challenge briefs: problem definition, anonymised user quotes and judging criteria. We can also join judging and post-event introductions.
How can developers take part?
Sign in to the company intelligence library to see the full evidence, the companies involved and the potential buyers, then pick a problem you want to solve. When you build something, we can introduce you to the companies struggling with it; capable developers can also build with us.
What does working with you look like, and how long does it take?
Hackathon-style delivery: ① understand the problem → ② compose modules from the agent library → ③ build the prototype together → ④ ship and iterate on real feedback. Usually two to four weeks from problem to a running prototype — a working system, not a white paper.

We are in an invitation-only period: apply through the waitlist, reviewed in batches every Wednesday at 09:00 UTC.
How is NeuronSpark related to MEMECMO? Do you still do GEO?
Yes. MEMECMO is the flagship GEO SaaS NeuronSpark built and productised: ten specialist agents covering ChatGPT, Gemini, Perplexity, Claude and Google AI Overview.

• Ready-to-use brand AI-visibility monitoring and optimisation → MEMECMO (app.memecmo.ai)
• Clients in the Hong Kong and Taiwan markets → the NeuronSpark GEO platform (geo.neuronsparkmedia.com)
• Measured readings on Chinese brands in Southeast Asia, plus our GEO method and selection Q&A → Southeast Asia Market Intelligence