Sean Lee | UX Designer | San Mateo, CA
Ask Rivi
Ask Rivi is a contextual AI assistant embedded within the recruiter workflow. It helps recruiters source and evaluate candidates, enabling them to take the right actions faster with confidence.
As a sole UX designer, I uncovered where recruiters need AI tool the most and designed the experience from the ground up, shifting Ask Rivi from a passive content generator to an active decision-support layer. Within a month of beta launch, 70% of controlled users were actively using the feature in their workflow.
The Problem:
Expanding AI Without a Strategy
Sutro had early success with an AI writing assistant (below). To capitalize on that momentum, leadership planned a rapid expansion: three separate AI tools for three separate tasks (Chat, Docs, and Sourcing).
The intent was valid but the execution lacked a cohesive UX strategy. Recruiters do not need more AI tools. They need AI that understands their context and workflow. Adding separate entry points introduced fragmentation:
-
Multiple tools mean multiple mental models and constant decision friction
-
Inconsistent behavior with different tool names and icons can reduce trust in AI outputs
-
No shared system led to duplicated logic and uneven quality
UX Strategy:
A Unified AI Experience
To address the fragmentation risk, my goal was to design a workflow-aware AI assistant as the global AI delivery model for the MVP, a cohesive AI experience that functions as a contextual task assistant rather than a collection of isolated LLM features.
To make that vision production-ready, I defined an interaction framework that standardizes how AI behaves from entry point to action completion, giving engineering a clear, consistent spec to build from and giving users a predictable, trustworthy experience no matter where they encounter AI in Sutro.
Through shadowing sessions and structured interviews with target users, I mapped the full recruiter workflow from job setup to offer. Sourcing stood out as the most manual, most time-intensive stage and the clearest opportunity for AI to deliver immediate value.
To move beyond ad-hoc AI features, I developed the AI Delivery Types framework - Global, Inline, and Passive models with specific AI capabilities tiled user mental models and workflow patterns.
While the North Star vision encompasses all three, we prioritized the Global delivery type for the MVP as the foundation of a unified AI experience with a consistent "command center" for AI interactions.
I designed the end-to-end AI workflow to shape the experience holistically from discovery and prompt input to processing, review, and iterative refinement. Rather than focusing on a single UI surface, this framework defines how AI behaves across each stage throughout the workflow.
Design Framework:
Contextual Sourcing Assistant
Rather than designing individual mockups for every AI interaction across Sutro, I built a reusable interaction model using sourcing as the primary use case (from UX strategy). The goal was to define how Ask Rivi thinks, responds, and behaves through one well-documented workflow, and hand that off to engineering as a blueprint they could apply consistently with the repeatable pattern and logics.
Brand Identity
I designed a distinct iconography for Ask Rivi to establish a unique AI identity within Sutro. The system scales from logo to action and status icons while remaining clear across sizes and color themes. The screenshot above details the design rationale and decision-making behind the iconography proposal.
Discover
Recruiters can easily launch Ask Rivi from the global header once they’re ready to source, making the assistant visible and accessible without disrupting the existing Sutro job flow. This entry point supports faster candidate discovery by introducing AI at the moment of need while keeping the experience non-intrusive.
Initial Launch
When recruiters launch Ask Rivi, the assistant recognizes the active job and current workflow state, guiding users toward the most relevant next steps in context.
Rather than open-ended prompts, the interface surfaces up to three guided quick actions ordered from soft to assertive, starting with exploratory suggestions like "show me recommended filters" before progressing to more committed actions like "apply filters and find candidates." This progressive trust model lets users ease into AI-assisted decisions at their own pace before they fully commit to relying on result driven by AI.
Input / Processing
AI latency is where user trust often breaks. To prevent the "did it work?" moment of doubt, I designed staged processing updates that surface in real time. By showing exactly what Rivi is retrieving, analyzing, and generating, the interface provides immediate feedback. This transparency reassures recruiters that their request is progressing, keeping them engaged instead of second-guessing the system.
Output / Review
Ask Rivi returns recommended filters with inline reasoning, citing the specific sources it used, styled as reference chips that open a document preview in a modal for instant verification without leaving the workflow.
Recruiters can also copy the output or leave direct feedback via thumbs up/down, feeding signal back into the system over time. The structured format is easy to scan, and actions follow the same progressive model: lighter confirmations first, stronger executions after.
Refine / Processing (Loop)
When a recruiter commits to an action, the interface enters a synchronized loading state to ensure a clean handoff between the user and the AI. Ask Rivi provides real-time status updates in the side panel while the main sourcing UI temporarily disables interactions, such as filters and buttons.
This interaction model prevents mid-execution conflicts and clearly signals that the system is processing. Once Rivi completes its analysis, control is handed back to the recruiter with results ready for review, making the transition from AI-processing to human-evaluation clear and deliberate.
Refine / Iterate
Once Ask Rivi completes its evaluation, control returns to the recruiter with a clear summary, breaking down matches into Strong, Medium, and Low tiers based on the applied filters and job criteria. The main UI updates simultaneously with ranked candidates and match strength indicators so recruiters can scan and act immediately.
Ask Rivi triggers actions and surfaces insights, but Sutro's main UI is where recruiters review, evaluate, and make final decisions. This separation ensures that the AI assistant remains a streamlined tool for workflow acceleration, while the Sutro platform remains the primary "source of truth" for deep exploration.
AI Informed - Humans Decides
Clicking a candidate card expands an inline detail view showing a Rivi Score, key highlights (experience, industry, location), and 3 to 4 fit signals grounded in job requirements and alignment documents. Two actions close the loop: "View Full Profile" opens the full candidate record in a new tab, while "Move to Pipeline" advances the candidate without breaking the sourcing flow.
Ask Rivi initiates, Sutro decides. Every design decision in this framework was made to keep the users in control while letting AI do the heavy lifting.
The Outcome:
Beta Launch and What Came Next
This was an intense 0 to 1 project moving fast with a small engineering team. From the beta rollout, over 70% of controlled users adopted the feature within the first month, with consistently positive feedback on performance and usability. Beyond the MVP, I worked with the product team to shape the vision for future releases - expanding Ask Rivi's capabilities across others parts of Sutro workflow, with adoption rate, frequency of use, and pipeline efficiency as the key metrics to measure progress.
