Conode.ai
Head of Design
Seven years ago I was employee #1, brought in to answer the messy early questions: how do we sell the vision, who are the first users, what’s the first win, and how do we make a knowledge graph platform feel powerful and intuitive—not punishing. We grew from a niche tool into a cross-industry platform for analysts and AI engineers. Started out as dRISK and evolved into Conode.ai
As Head of Design at a start-up, there's no management; what matters is flexibility in process, in the face of the beautiful chaos of many 0 to 1 initiatives. Our 4-person team designed, strategised, and shipped almost everything you see here. We drive Conode’s product design direction and brand evolution.
-
We first needed to sell the vision
-
Then drive grant and VC fundraising by translating a deeply technical platform into a clear, compelling narrative
-
Flag early-adopter industries and users are best positioned to adopt the hyper-technical software
Why did the founder bring in a designer as employee number 1?
-
To scale design strategy from pre-seed through the growth stage, adapting design focus at each phase
-
Communication for high-stakes audiences: investors, grant committees, and enterprise decision-makers
-
To build a user-centered foundation by validating ideal experiences before or while engineering
-
To work with product and engineering to turn ideal design vision into executable roadmaps—not just unbuildable prototypes
No idea what a knowledge graph is?
I gotchu...
Imagine being able to take many different kinds of data—spreadsheets, documents, free-text reports, emails, images, video feeds, even messy or mislabeled files—and bring them all into one place.
A knowledge graph is an environment designed to do exactly that. Instead of forcing data into rigid formats, it connects information based on how things relate to each other.
This makes it possible to explore, analyze, and ask questions across all your data at once—even when it comes from different sources or doesn’t neatly line up.
The result isn’t just more data in one system—it’s clearer insight from seeing everything together.
This is what I started off with
...right?
Extremely Powerful
...Sure
-
Visually driven knowledge-graph builder — think Photoshop for knowledge graphs
-
Massive scale — capable of handling hundreds of millions of nodes and edges
-
Schema-free by design — taxonomies can evolve as new insights emerge
-
In-memory graph architecture — ideal for rapid exploration and graph learning
-
Built for speed — optimized for fast data ingestion, iteration, and experimentation
Can only be used by
the indoctrinated
-
High barrier to entry — required deep expertise in knowledge graphs to use effectively
-
Limited system feedback — users had little visibility into system state or progress
-
Founder-centric design origins — initially built to serve the creator’s workflow, as part of his PhD. rather than a broad user base
-
Local-only deployment — not cloud-native in its early form
-
Blank-page problem — users were often unsure where to begin or what to do first
Who is the user, and what do they want?
-
Industry entry point: Secured major grant funding to build a knowledge graph of everything that could go wrong on the world’s roads—used to test and train autonomous vehicles in simulation before real-world deployment
-
Early adoption: Leveraged the team’s network to place the tool quickly into the hands of AI engineers at autonomous-vehicle companies
-
Rapid learning loop: Iterated fast through real-world usage—direct observation, follow-up surveys, customer interviews, and structured workflow assessments
-
Comparative insight: Studied the tools customers already relied on and compared them directly to ours, exposing gaps between expert-grade capability and everyday usability
-
Key finding: While powerful for AI engineers, the platform didn’t scale across organizations—regulators and executives needed insight without deep technical overhead
-
Observed friction: Even expert users struggled with simple tasks, revealing usability breakdowns that limited broader adoption
-
Design response: Evolved the platform toward more familiar layouts and interaction patterns—designing for what users already knew and felt comfortable with
-
Outcome: Created a foundation for design decisions informing scaling of the platform, beyond specialists to cross-functional teams and decision-makers
Expansion, streamlining and growth baby! woooo! - Conode.ai
-
Customer insight: Customers needed the platform to fit with their existing workflows—not force new ones.
-
Research & experimentation: Conducted usability studies, observed real-world workflows, and ran rapid low/mid-fidelity experiments to identify friction points.
-
AV impact: Drove a 3× increase in users within autonomous-vehicle companies, expanding adoption beyond testing and training into scenario management.
-
Cross-industry expansion: The workflow-level design translated naturally to analysts and data scientists across insurance, logistics, banking fraud, retail, grant organizations, and food & beverage.
dRISK - Conode
brand evolution
As we expanded beyond autonomous vehicles, we needed a brand that reflected what we truly were: a flexible knowledge-graph platform, not a single-industry tool.
I defined the brand strategy and led in-house efforts in creating the vision—positioning, visual language, and messaging. This shift enabled Conode to engage multiple verticals in parallel, launch targeted pilot programs, and validate which use cases were worth building for.
The LLM boom
The LLM explosion was a great catalyst in reshaping how one interacts with Knowledge Graphs, and brought up fascinating challenges.
-
How do you create, curate, and evolve a knowledge graph through conversation—not overwhelming visuals?
-
How do you verify new data has actually been incorporated without errors?
-
How do you check whether an answer from a GraphRAG system is true?
-
And can static interfaces really support this level of flexibility?
Time to rethink the entire interaction model.
-
Designed flexibility into the system by mapping the actual steps users take to reach a meaningful insight
-
Rebuilt workflows around agents, not screens—matching how people think, not how graphs are structured
-
Introduced guided entry points: onboarding pages, tutorials, tooltips, and contextual help
We designed and built a custom GraphRAG, front-facing agent on top of the knowledge graph—dramatically lowering the barrier to entry.
What actually changed?
-
Designed explicitly for the “I’m completely lost—what can I do?” moment
-
'Designed' LLM behavior to act as a coach, not just a query engine
-
Built purpose-specific agents that walked users through ingestion, cleaning and presentation of data
-
Designed explicitly for the “I’m completely lost—what can I do?” moment
-
'Designed' LLM behavior to act as a coach, not just a query engine
-
Built purpose-specific agents that walked users through ingestion, cleaning and presentation of data
Using agents to do most users' tasks dramatically increased usability—and, when paired with bespoke front ends, opened up an entirely new market: GraphRAG for small and medium-sized businesses.
These users don’t have enterprise-scale data problems—but OpenAI isn’t knocking on the door of your neighborhood burger joint to sell them a corporate contract either.
So we built a self-serve platform that let non-technical teams create, customize, and evolve their own agents—without needing a PhD in graphs or machine learning.
The shift from building graphs by Agents to accessing them by Agents.
mmm agentic...
How it works
(no lab coat needed)
-
Upload your data, buid the graph
-
Manage permissions - who needs access to which bit of the graph?
-
Map how those agents are actually used
-
Feed insights back into Conode
-
Identify patterns of voice
-
Iterate on LLM behavior
Same underlying power. Far less terror.
Repeat, but it bears mentioning...
What’s the impact?
In 2018, we had no users, no customers, and no VC funding.
By 2026, we secured 2 VC funding rounds, won several rounds of grants, and gained global customer traction.
What began as software for autonomous-vehicle testing evolved into a platform used across logistics, insurance, and SMBs—enabling teams to deploy domain-specific LLMs grounded in their own knowledge graphs.
We’ve successfully exited. I’m now looking for the next challenge.
Copyright Kiran Jesudasan 2026