Inspiration
My teammate and I have both watched ideas die quietly. I had an idea to build a solution powered by AI for plants but that died. This was not from a lack of ambition, but from the gap between "I have an idea" and "I know what to do next." Most planning tools either oversimplify, handing you a generic pros-and-cons list, or overwhelm you with information you don't know how to act on.
What stuck with us was a specific pattern: people don't usually fail because their idea was bad. They fail because they never stress-tested the assumptions hiding inside it, and they had no system holding them accountable once the initial motivation faded. We wanted to build something that did both , challenge the thinking before the plan exists, and stay involved after the plan is made, not disappear the moment it's generated.
What it does
LaunchPad is an AI co-founder for the zero-to-one stage of building something. It takes a user from a vague, half-formed idea to a reality-checked execution plan, then holds them accountable to actually following through.
The flow runs in five stages. First, the user describes their idea in plain language, no forms, no dropdowns. Second, Launchpad interrogates the idea, surfacing 3-4 critical assumptions buried inside it and asking pointed, idea-specific questions rather than generic ones. Third, it maps real constraints: available time, resources, and the single factor most likely to derail the project. Fourth, it produces a structured output, a sharpened one-paragraph description of the idea, the top assumptions to validate with concrete validation methods, a 30/60/90-day milestone plan, one action to take in the next 48 hours, and one honest risk worth naming.
The fifth stage is where Launchpad becomes more than a planning tool. With the user's permission, it syncs milestones to their Google Calendar as real commitments, checks in partway through each milestone window, and, critically, doesn't unlock the next milestone if the current one isn't met. Instead, it asks what happened and adapts the plan based on the reason, rather than letting the user quietly drift off track. Throughout, Launchpad is explicit that it's a thinking tool, not a verdict. It never tells a user whether their idea is good. That judgment call stays with the human.
How we built it
We split the build along our strengths. Samuel built the web platform the conversational interface for idea intake, the interactive milestone dashboard, and the Google Calendar integration. He also focused on the reasoning architecture: designing the prompt framework that forces the AI through the assumption-interrogation stage before it's allowed to generate any plan, so the output feels earned rather than templated. I focused on putting together the pitch video and getting our solution ready for submission. The core technical challenge was making the AI's questions feel specific to this idea rather than generic startup-advice boilerplate, we engineered the system prompt to extract and reference the user's own language back at them, rather than falling into a fixed question bank.
Challenges we ran into
Getting the AI's interrogation stage to feel genuinely sharp, rather than a generic "have you validated your assumptions?" , took real iteration. Early versions of our prompt produced questions that could apply to almost any idea, which defeated the entire premise. We had to engineer the prompt to pull specific nouns and claims out of the user's own description and turn those into pointed challenges, rather than relying on a generic question template. Working within a one-week window also meant constant scope discipline. The milestone re-evaluation logic (adapting timelines based on why a user missed a deadline, not just that they missed it) was the feature most tempting to cut under time pressure, and the one we fought hardest to protect, because it's the core differentiator between Launchpad and a static plan generator.
Accomplishments that we're proud of
We're proud that Launchpad doesn't just generate a plan, it earns the plan through a structured reasoning process the user actually experiences, and then stays present long after most tools would have walked away. The milestone-gating system, where progress is genuinely contingent on follow-through rather than a list the user can ignore, is the piece we think most directly tackles the real reason ideas stall. We're also proud of how deliberately we built in the human-in-the-loop boundary. It would have been easy to let the AI render a verdict on whether an idea is "good" , we resisted that, because we believe that judgment should never belong to the model.
What we learned
Building Launchpad sharpened our understanding of where AI reasoning genuinely adds value versus where it's just generating plausible-sounding text. The assumption-interrogation stage only works because it's grounded in the user's specific language, a lesson in how much prompt design matters for making AI output feel earned rather than templated. We also learned, somewhat personally, what Launchpad itself is trying to teach: that good execution requires structure and accountability, not just a good idea. Building this under a one-week deadline was its own lesson in exactly the thing we built the tool to solve.
What's next for Launchpad
Near-term, we want to expand milestone re-evaluation to recognize patterns across missed deadlines, if a user consistently struggles with the same type of milestone, Launchpad should surface that as a structural insight, not just adjust one timeline. Longer-term, we're interested in collaborative mode for small teams building together, shared milestone visibility, and integrations beyond Google Calendar, Notion, Slack reminders, or task managers people already use.
Log in or sign up for Devpost to join the conversation.