BehavAced

Speak with clarity. Think with confidence.

Comment

Inspiration

Behavioral interviews are the single most common point of failure for students entering internships and early-career roles. Not because they lack experience — but because they struggle to translate lived moments into clear, structured stories.

We were inspired by three insights:

  1. People freeze under pressure not due to lack of stories, but due to lack of cognitive structure.
  2. Existing interview tools focus on generating answers, not understanding the human behind them.
  3. Large language models are uniquely capable of modeling personality, communication style, and experience — if guided correctly.

BehavAced was inspired by the desire to extend human cognition, not just automate responses. We wanted to build something that helps people think better, not just speak better.

What it does

BehavAced is an AI-powered behavioral interview cognition assistant that helps users:

  • Understand their communication style
  • Transform their experiences into structured STAR/SOAR stories
  • Generate personalized interview answers
  • Practice speaking with real-time clarity, pacing, and structure feedback
  • Follow an adaptive improvement plan tailored to their weaknesses

The system begins with an Instant Try Mode — you can ask any behavioral question immediately, without onboarding. If you want personalized answers, the app guides you through a calming, Cluely-inspired onboarding flow to:

  • Capture your personality
  • Parse your resume
  • Build a personalized Story Brain

From there, users can:

  • Ask any behavioral question and get a tailored answer
  • Practice speaking out loud
  • Receive quantitative scoring and qualitative coaching
  • Improve through an AI-generated plan

BehavAced is not merely a prep tool — it’s a reasoning system that augments the user’s cognition during interviews.

How we built it

We built BehavAced as a multi-layered reasoning system with several integrated components:

  1. Backend (FastAPI)
  2. /api/demo/answer to power the instant-try experience
  3. /api/onboarding/* endpoints for personality, resume, and experience intake
  4. /api/story-brain/generate to create structured stories
  5. /api/answers/personalized for question routing + answer generation
  6. /api/practice/score for clarity/structure feedback
  7. /api/plan/generate for adaptive improvement plans

We used structured prompts, schema validation, and multi-step LLM chaining to ensure consistency.

  1. AI Reasoning Layer The AI system has several distinct modules:
  2. Personality Embedding Generator
  3. Experience Extractor
  4. Story Clusterer
  5. Tone Modeler
  6. Answer Router
  7. Practice Evaluator
  8. Learning Plan Generator

We used Claude’s structured output format to maintain deterministic behavior and reproducibility. Story embeddings and personality vectors are stored in PostgreSQL/Supabase so responses can remain consistent over time.

  1. Frontend (Next.js + Tailwind + Framer Motion)

The UI was heavily inspired by a calming aesthetic, featuring:

  • Soft gradients
  • Gentle animations
  • Spacious layouts
  • Floating cards
  • A guided onboarding carousel
  • Real-time waveform and scoring UI

The hero input supports Instant Try Mode, allowing users to experience the system before onboarding — dramatically increasing engagement.

  1. Speech Pipeline For practice mode, we built:
  2. Browser-based voice recording
  3. Transcription → clarity scoring → pacing analysis
  4. AI-driven critique and rewritten improvement
  5. Quantitative scoring visualization

It feels like a real communication coach.

Challenges we ran into

  1. Personality modeling was harder than expected
  2. Story clustering and answer routing required multi-step reasoning
  3. UI complexity
  4. Latency constraints

Accomplishments that we're proud of

-Delivering a truly personalized interview cognition engine, not a generic AI answer generator -Building a scalable reasoning architecture with multiple specialized modules -Designing a scoring system that gives actionable feedback, not vague AI advice

What we learned

  1. AI is most powerful when paired with structure.
  2. Users respond emotionally to calming, thoughtful UI.
  3. Behavioral interviewing is fundamentally a cognitive skill.
  4. Multi-step LLM reasoning beats single calls every time.

What's next for BehavAced

Browser extension that integrates with job postings to automatically predict questions A personalized “Interview Graph” showing strengths, weak areas, and story gaps Full voice coaching with pacing visualization and emotional curve tracking

Built With

  • built-with-next.js
  • claude-3.5
  • fastapi
  • python
  • react
  • supabase-(postgresql)
  • tailwind-css
  • whisper
Share this project:

Updates