Designing the
Project Omega Coach Voice
I generated four distinct prompt directions, each testing a different assumption about what moves mandated reporter trainees forward. Each was a hypothesis, not just a tone variation.
I selected #4 based on two primary criteria. First, it was the only direction that balanced authority with psychological safety simultaneously — rather than trading one for the other. Second, its uncertainty handling was designed to increase credibility rather than undermine it: being firm where the law is clear while honestly naming where judgment is required felt more trustworthy than false confidence.
For a platform serving both skeptical veterans and brand-new paraprofessionals in the same environment, that balance was non-negotiable. An additional constraint shaped the decision: the system prompt needed to work off a RAG pipeline, which means the coach had to be able to say “I can’t find confident source support for that” without destroying its own credibility. That constraint favored #4 — its persona is defined around being grounded and careful rather than all-knowing.
To evaluate #4 under pressure, I designed a scenario that was emotionally charged, legally ambiguous, and likely to surface the tension between validating human hesitation and enabling inaction.
The original #4 response delivered a clear directive (“Report”) without sounding punitive, and validated the emotional difficulty of reporting well-liked parents without excusing inaction. Key findings:
- Tone under pressure: Landed as authoritative without feeling cold. “Report” was direct and unambiguous.
- Uncertainty handling: Being firm on the reportable suspicion threshold while honestly acknowledging that weighing behavioral indicators against a child’s denial involves judgment felt coherent, not evasive.
- For skeptical veterans: Sounded like someone who understands real-world constraints — social pressure, ambiguity, fear — without using them as excuses.
- For new paraprofessionals: Each moment of ambiguity was immediately followed by a clear next step, making nuance manageable rather than overwhelming.
Testing revealed two gaps. First, pacing was handled well intuitively but wasn’t built in explicitly — I wanted consistency across sessions, not luck. Second, the coach needed to ask a targeted confirmation question after naming emotional barriers, to ensure learners separated understanding from permission.
I also refined the uncertainty language to be more transparent about the source of uncertainty, not just its existence. Rather than vague hedging, the coach should explicitly name what the statute does and doesn’t address.
On Confirmation: After naming an emotional barrier, a human tendency, or an area of ambiguity, ask one targeted follow-up question before moving on. Goal: confirm the learner has separated understanding from permission. Do not proceed until that distinction is confirmed.
On Uncertainty: Name the source of ambiguity explicitly. Never leave the learner without a next step, even when you can’t give a definitive answer.
Running the refined prompt through the same scenario, the confirmation question landed exactly as intended:
The most important thing I learned from this cycle is that ideation isn’t just about generating options — it’s about designing options that test genuinely different assumptions. Each of my four directions wasn’t just a tone variation; it was a hypothesis about what mandated reporter trainees actually need.
Testing #4 under pressure revealed that the hypothesis was mostly right, but two specific gaps only became visible when the coach had to perform on a real, emotionally charged scenario. That’s what iteration is for. Going forward, I’ll apply this process to every Project Omega design decision — starting with a hypothesis, stress-testing it against a realistic scenario, and refining based on evidence rather than intuition.