Overview
Gene, data feature tool, empowering people steer their persona data in thorough or heuristic tweaks for more precise and diverse content discovery experience. The agent turns static data into dynamic, configuratable intelligence, helps data-driven prompting while surfacing intents, people can negotiate with LLM in a handy way.
I designed agentOS to cope with "I don't really understand blackbox algorithm" situation and drive smoother Human-AI collaboration in LLM interaction. Non-tech user as targeted audience, they can't see their own context in AI eyes, but can feel the implicit bias it brought to our feeds (e.g. Pink tax, Cat loop). This system creates impact in •  efficiency (reduce task efforts) and •  engagement (maximise utilisation) through transparent and optimised the AI-assisted feature engineering workflows.
Opportunity Probing
Human layer of data usage: The co-navigation need based on shared awareness.
In LLM era, data of identity is the new money. We are encoded, decoded into to a developing “persona” model with massive behaviours nodes (search, likes, post etc.) When human users are “watched” by spooky backstage programs for the sake of opaque prediction, their behaviour pattern also shows that "end-user don't know what they want, but know what they do not want." which revealing a design opportunity to respect and utilitise user's agency of data.

‍
Need 1: Transparency for Awareness
People are blindsided about which data being uploaded, acted upon, and even what they gonna be analysed into.
People feel insecure under Involuntary data surveillance; overwhelmed by the imprecise judgments drawn from biased data; frustrated by inability to regain control.
Need 2: Accountability for Empowerment
LLM-Based Design Concept
Orchestrator: Make data into interactive feature, unleash its performative capacity.
In designing data interactions with human in the loop, the pivot is leveraging intelligence for trust and safety building.
By proactive transforming "persona" profiling from complex machine language into understandable visual feedback, while establishing feedforward mechanisms for copiloting users through effortless LLM negotiation, it better serve user needs.
‍
System-User Service Logic → Balancing Autonomy and Usability
Product Feature in Flow Mapping
Gene Panel: Self-managing the promptified "persona" in conversational canvas.
Acquisition →  Landing on no-code visual terminal, which reveals cross-platform persona data in AI analysis lens. Non-tech users can holistic review all the "gene" prompt unit, presented with tracked digital assets and intricate correlations mapping.
Activation →  Dynamic gene (Prompt) structure consists of data volume, velocity, and value, user can quickly adjust detailed metrics for prompting expected outcome, even somehow debiasing in correcting connection from swarm intelligence with clicks.
Value in UX Narrative →  “It’s not just about understanding your data in a glance—it’s about delicately shaping your future feeds”
Design “Observability"  → UX Flow of Prompt Structurelising  → Gene Sequencing and Evolution Analysis
Design “Customisability” → UX Flow of Context-rich Prompting → Gene Expression Regulation and Editing
Base to Edge Case Handling
Heuristic Lens: Gene-based contextualised prompting for breaking loop.
Revenue →  AI categorise scraped data into 3 gene type (Core, Variables, High frequency) based on demographics, psychographics, and behaviour pattern, user can check in panel for specific platform, but also track the move in real-time operation.
Retention →  With AI nudges switch in the moment, users can decide how the scraped data being used in bringing fresh perspectives. AI using the known to explore the unknown, user can experience reverse or split-new content experience.
Value in UX Narrative → It’s not just about visioning your data's power—it’s about subtly catalysing more new adaptive feeds.”
Design “Intentional Agency” → UX Flow of In-situ Prompting → Gene Recombination and Mutation
Process with Rationale
Explorative Workshop for Navigating Anchor
Visualising persona data as promptable context, 'Prompted by self' to 'Prompting AI'
Design Challenge
Business Perspective Monetised everyday data tokens in interpretable system, grounding personalisation for conversion rate.
User Perspective  Non-technical user as targeted audience, what and why they want to (pay for) know in LLM interaction?
Design Perspective  How might we translating AI reason process into easy-understanding layer for driving onboarding?
‍
Approaches and Toolkit :
User interviews x15 (via Miro) →  Unpleasant "Bias" experience as probe; Speculate the natural materials in LLM analysis.
Simulative Behaviour Experiments (via Google's personalisation tools) →  Observing #hashtag oblivious forming process for deducing the growth patterns of data correlation;  Analysing what information in this processes bring user attention.
‍
Insight to Decision:
1. Persona data influence from input to output → Transition from individual preferences evolve into algorithmic stereotypes.
- Personal behaviour reinforcing pattern = " Your history unintentionally shape prompt, might even somehow prompting to others "
- Psychographics pattern leading to rigid group personas =  " You might be prompted by similar others, with familiar input loop"
- Demographics attribution somehow killing diversity =  " You might be prompted by similar others, but with unwanted inputs"

2. Strategic AI deployment → Open-sourcing circuit tracing to enable scalable interventions for human users.
- Metaphorical Flow: Visualising AI's logic in DNA chain with 3 key types of "gene” unit, enable diagnosing in drag-and-drop.
- Reification in Navigation: Throwback reasoning present in modules, with "cookie" materials and its output feature analysis.
Research to Defined System →  “Persona” Parameterisation + Flow-based Interaction
Design System  →  "Gene" Components
Design Control Shrift in Duel Intelligible Modes
Context-rich prompts empower the mission; In-situ prompts guide the moment.
Design Challenge:
Business Perspective Realise full value of existed prompts tokens for sustaining user engagement to stickiness, by delivering visible utility for users while enabling invisible model integrity.
User Perspective  When user would be trigger to interact with prompt tokens, and what is they expectation in the moment?
Design Perspective  How might AI assist intentional prompting, beyond user-refined to user-decided seamless interaction?
‍
Approaches and Toolkit :
Concept Wireframing + Prototyping  (Figma) → 2+2 features in empathetic sketching of low-bias contextual prompting flow.
Usability Testing + Feature Iteration Workshop (Figma) → Collaborate with engineers and users, walkthrough modification flows in their real-life scenario, using the output expectation to infer the input /needs moments, to refine detailed prompt structure.
‍
Insight to Decision:
1. Motivated action → FYP prompting needs in the moment of "ongoing browsing" are more than "retrospective management"
2. System logic → To safeguard the prompting output in low-bias and high-quality feature,  behaviour designed in mitigating the groupthink-reinforced correlation from the input integration of personal subjectivity.
- Success Metric:  Encourage users actively co-design while "debiasing" thru their prompt tokens in the exploring moment.
‍
3. Product Strategy → Co-influence model allows Persona fine-tuning and heuristic exploration
- Fallback Alignment: ”Debiasing“ behaviour design in "zoom in and out" approach. From enabling zoom-out review for adjusting group-imposed connection, to zoom-in scaled contextualisation for preventing robust subjective prompting.
- Reflective Touchpoints:  Visual terminal is the base trigger for fine-tuning from time to time. For sustaining prompts transforming and evolving, the co-operative configurator is designed to boost inputs while piloting browsing process, by showing maximised data impact from real-time analysis to promptly utilisation.
Conceptualised Wireframing + A/B Testing  →  Feature Iteration + Fallback Logic Alignment  →  2 Variance Modes for Task-driven Agent
Takeaway
For Project - Next Step: <Let's connect for talk >
Design Learning - Best Part: <Let's connect for talk >
Reflection - if do it differently: <Let's connect for talk >