Hira

Hira, the end-to-end AI-powered technical interview platform designed to streamline hiring

Comment

💡 Inspiration We watched too many brilliant engineers bomb interviews—not because they couldn't code, but because they froze up under pressure. 🥶 Traditional coaching costs $200/hr, and asking friends for mock interviews is awkward. We realized the world didn't need another LeetCode clone; it needed a simulation. We built Hira to democratize "Senior Engineer" level mentorship, making elite interview prep accessible to everyone, 24/7. 🌍✨

🚀 What it does Hira isn't just a chatbot; it's a full-blown Interview Simulator. 🎙️💻 It acts as a strict-but-fair Senior Technical Interviewer who can see you (via camera), hear you (via voice), and critique your code in real-time. It detects if you're cheating, pushes you to explain your logic, and parses your actual resume to ask personalized questions. It's the closest thing to a real FAANG interview without the anxiety. 🔥

⚙️ How we built it We engineered a high-performance "Agentic Sandwich":

The Brain: Google Gemini 1.5 Flash handles the heavy lifting—multimodal reasoning, code analysis, and resume parsing. 🧠 The Voice: ElevenLabs Conversational AI brings Hira to life with ultra-realistic, interruptible speech that laughs and breathes. 🗣️ The Backbone: A robust Go (Golang) backend for blazing-fast concurrency, serving a slick Next.js 14 frontend styled with Glassmorphism. ⚡ The Eyes: Face-api.js runs client-side computer vision to track eye contact and detect multiple people. 👀 🚧 Challenges we ran into The Latency War was real. ⚔️ Making an AI "think," check for cheaters, analyze code, and speak back in under 2 seconds felt impossible. We had to optimize every millisecond, using optimistic UI updates and parallel processing in Go to ensure Hira never felt "robotic" or laggy. Also, keeping the AI from hallucinating a "perfect score" when the code was broken required some serious prompt engineering gymnastics! 🤸‍♂️

🏆 Accomplishments that we're proud of We successfully "hacked" the LLM's senses! 🤯 By converting visual signals (like looking away from the screen) into dynamic "System Text Alerts" injected into the chat history, we tricked a text-based model into "seeing" physical behavior. Now, Hira can verbally reprimand you for bad eye contact or suspicious movement—feeling eerily human without needing expensive video processing. 🕵️‍♂️✨

🧠 What we learned We learned that Prompt Engineering is actually rigorous Logic Design. 📐 It’s not just writing English; it’s building complex state machines where Voice, Vision, and Code Analysis agents have to coordinate perfectly via JSON. We also discovered that Voice is the new UI—when an AI speaks with emotion, users instantly treat it with more respect and focus than a text box. 🗣️💬

🔮 What's next for Hira We’re taking Hira to the next level! 🚀

🏢 Company Modes: Unlock specific interview styles (e.g., "Google Mode" for graphs, "Amazon Mode" for leadership principles). ⚔️ Multiplayer Mock: Peer-to-peer interviews where Hira acts as the neutral judge and moderator. 📈 Long-term Coaching: Tracking your improvement over weeks to predict exactly when you're ready to apply.

Built With

Share this project:

Updates