Knowledge Agent
Altifypedia
A single source of truth trained on 600 product docs, with role-tailored answers for sellers and engineers.
/ The Problem
User testing at Altify was bottlenecked by the time it took to run, transcribe, and synthesise sessions making it a rare event rather than a continuous practice.
Running a moderated user testing session requires a facilitator, a note-taker, a scheduler, and hours of post-session synthesis. Most product teams simply couldn't afford the overhead, so decisions were made without real user input. When testing did happen, weeks could pass before findings reached the team in a usable form.
Pain points
- Sessions required 2–3 people to run and produced raw recordings that took hours to process.
- Insights arrived too late to influence the decision they were meant to inform.
/ The Solution
I designed a system of 5 AI agents that replace the need to recruit and schedule real participants — giving the design team on-demand access to synthetic users who behave, react, and push back exactly like Altify's core roles.
Instead of waiting weeks to coordinate sessions with real sellers, the team can now run a test at any point in the design process. Three synthetic user agents — a CRO, a VP Sales, and a Senior Enterprise AE — each read a prototype or flow and respond in character: thinking aloud, flagging friction, asking the questions their real counterparts would actually ask. A fourth agent writes the test scripts so every session is consistent and research-quality from the start. A fifth synthesis agent then compares all three perspectives, surfaces where they agree and diverge, and produces a single actionable recommendation the team can act on immediately.
/ How It Works
1
Script Writer Agent
The designer shares a brief, prototype link, or feature description. The Script Writer turns it into a moderator-ready test script — structured sections, neutral tasks, probing guides — so every session asks the same things in the same order and findings stay comparable.
2
Synthetic User Agents × 3
Three AI personas representing Altify's core user types run the session: Cristina Jones (CRO), John O'Connell (VP Sales), and Morgan Hale (Senior Enterprise AE). Each thinks aloud, reacts to what's shown, flags confusion and friction, and gives honest role-specific feedback — never softening criticism to please the designer.
3
Comparison Agent
Once all three personas have responded, the Comparison Agent takes their answers side by side. It identifies where all three agree, where they diverge and why, what none of them addressed, and produces a synthesized recommendation with a confidence rating — so the team gets one clear direction instead of three opinions to reconcile manually.
/ Impact
User testing went from a rare, resource-intensive event to something teams could run continuously with minimal overhead.
3×
Faster synthesis
1
Person to run a session
<5min
Report turnaround
Key outcomes
- Teams run testing rounds without a dedicated researcher present.
- Findings are available before the end-of-day standup instead of days later.
- Consistent probing patterns make cross-session comparison reliable for the first time.
- More sessions run per sprint leading to more decisions grounded in actual user behaviour.