Aptitude AI

Inspiration

Students graduate without the soft skills employers demand, while campus career centers remain overwhelmed and under-resourced. First-generation students and anxious achievers often lack access to personalized coaching. We are deeply passionate about alleviating the intense anxiety and imposter syndrome that paralyze so many capable young adults today. By providing immediate coaching without judgment and verifiable digital credentials for skill mastery, Aptitude AI bypasses the bottleneck of traditional advising. Our core mission is to democratize career preparation and empower every student to confidently step into the professional workforce. This commitment is profoundly meaningful because it transforms a terrifying life transition into an opportunity for equitable growth and ensures higher education delivers on its ultimate promise for everyone.

What it does

Aptitude AI is a career companion that remembers the user instead of acting as a generic chatbot. It directly addresses a severe expectation gap where a large majority of students feel prepared for the workforce, but most employers find them lacking in soft skills. The platform offers the specialization of a premium coach with the low-friction accessibility of a mobile application. It captures the market whitespace by specifically targeting the murky middle demographic of anxious sophomores and juniors. The application features an opportunities feed with evidence-based Confidence Dossiers that highlight user readiness and provide Socratic prompts for independent decision making. It also includes a resume translator that converts academic experiences into applicant tracking system-friendly and NACE aligned bullet points.

How we built it

We built a Next.js App Router mobile web application using React 18 and TypeScript with Tailwind CSS for styling. We use NextAuth for identity management alongside Postgres on Neon with pgvector for long-term memory. We run GPT 4o mini after each mock interview to extract structured insights and generate embeddings with text embedding 3 small before storing them in our database. The coach agent uses a query tool with a time decay vector search to ensure replies reference and retest prior weaknesses. We inject dynamic context like target weaknesses and user profiles into mock interviews and dossiers. OpenAI Realtime powers our live voice feature, and we maintain a fallback path utilizing Whisper and text-to-speech. We implemented consent-gated analytics with PostHog and added a cookie-based demo mode for hackathon presentations.

Challenges we ran into

• We designed a complex schema and time decay formula to ensure the coach receives relevant evidence without overwhelming the context window.
• We integrated OpenAI Realtime voice features while maintaining consistent coaching instructions and building a reliable fallback path.
• We engineered the application to degrade gracefully and use local storage when the primary database is disconnected to ensure users always receive a seamless experience.
• We refined the product focus to ensure the feedback and dossiers remain clear and actionable instead of providing generic advice.
• We balanced the scope of building mock interviews and resume translation alongside the opportunities feed to prevent feature creep.

Accomplishments that we're proud of

• We aligned our platform with UN Sustainable Development Goal 4 for Quality Education and SDG 8 for Decent Work and Economic Growth by democratizing career preparation.
• We developed a commercially viable strategy that utilizes a free consumer app as an acquisition channel to aggregate engagement data.
• We positioned the product to secure enterprise contracts with universities by anchoring our pricing to the tuition revenue saved by preventing student dropouts.
• We built a highly robust memory-augmented AI system that truly remembers the student and adapts to their personal growth trajectory over time.

What we learned

• We learned how to design a memory-augmented AI product where retrieval uses an exponential time decay formula to weigh recent evidence more heavily.
• We structured episodic storage for session logs and semantic extraction for NACE aligned competencies.
• We implemented vector search with time decay so the coach can reference past struggles and suggest specific practice areas.
• We integrated OpenAI Realtime for live voice capabilities while maintaining the same coaching logic.
• We learned to degrade gracefully when our database is unavailable by using local storage so the application still delivers value.

What's next for Aptitude AI

• We plan to deepen our integration with institutional learning management systems like Canvas or Blackboard.
• We will expand our micro influencer marketing campaigns on platforms like TikTok and Instagram to organically reach our target demographic.
• We aim to partner with early adopter university career centers to launch pilot programs and gather direct institutional feedback.
• We intend to refine our quantitative assessment modules to generate even more robust digital badges for student portfolios.

Built With

• eslint
• gpt-4o-mini
• next.js-14
• nextauth
• openai
• openai-(gpt-4o-mini
• pgvector
• playwright
• postgres-(neon)
• posthog
• react-18
• realtime
• tailwind-css
• text-embedding-3-small
• text-embedding-3-small)
• tts
• typescript
• vercel
• whisper