Where developers grow together

interview-pilot

🚀 Excited to share a project I’ve been building — InterviewPilot, a real-time AI mock interview agent for Android developers.This started with a simple idea:What if an AI interviewer could actually conduct a technical interview through voice, instead of just chatting through text?So I started building it.🎙️ What InterviewPilot can do:• Real-time voice conversation with an AI interviewer• Android-focused technical interviews• Senior/Lead Android Developer interview scenarios• Adaptive follow-up questions based on responses• Interview state management• Answer evaluation and competency scoring• Kotlin + Jetpack Compose Android app• Agora RTC + Conversational AI• Python + FastAPI backendThe interesting part for me isn't just connecting an LLM to an Android app.I'm exploring how to build an actual AI interview agent that can:→ Understand a candidate's response→ Decide when to ask follow-up questions→ Adjust interview difficulty→ Maintain interview context→ Evaluate answers→ Move through different interview stages→ Generate a structured interview reportCurrent architecture:Android App↓Agora RTC↓Agora Conversational AI↓Interview Agent↓Interview State + Question Engine + Evaluation Engine↓FastAPI BackendThe first milestone is now working end-to-end — I can have a real-time voice conversation with the AI interviewer. 🎙️This is still a work in progress. Next, I’m planning to explore things like RAG, resume-aware interviews, interview history, deeper evaluation, observability and production deployment. 🔗 GitHub: https://github.com/panky1/interview-pilot

TourBuddyAI

TourBuddy AI is a voice-based travel companion that I built using SwiftUI and Agora Conversational AI. It lets users simply tap, speak, and ask questions about their travel destination instead of typing and searching. For example, a traveller can ask, “Tell me about Mysore Palace,” “What can I visit nearby?” or “What should I explore next?” and have a natural voice conversation with the AI. I built the SwiftUI mobile interface, TourBuddy branding, tap-to-talk experience, AI orb, and integrated Agora’s real-time voice agent so the user can speak with the AI and receive spoken responses. How it helps: TourBuddy makes getting travel information hands-free, quick, and conversational, which can be useful while sightseeing, walking, or planning what to explore next. The concept can later be expanded with maps, live location, routes, attractions, and personalized recommendations.

Sahara | The Voice for Complete Indic Patient Care

Sahara A Multilingual Voice AI Ecosystem for Elderly Patients & Everyday Healthcare Sahara is a multilingual voice AI ecosystem for elderly patients and everyday users. It works as a daily health helper and transforms into an emergency lifeline the moment something goes wrong. The patient only has to talk. If they can't, a family member or anyone nearby can take over. Built for the Agora Voice AI Hackathon 2026, Sahara was conceptualized, designed, and built in just 3 days. THE PROBLEM Many elderly people in India live with high blood pressure, diabetes, or heart conditions and spend long hours alone. During a heart attack, stroke, or fall: They may not be able to unlock a phone. They may not be able to type. Family members may find out too late. When help finally arrives, nobody may know the patient's medical history. The most important first hour can be wasted. On normal days, there are smaller but equally important problems: Forgetting medicines Difficulty using English-first health apps Language barriers Overpaying for branded medicines Not knowing which government health schemes they are eligible for Healthcare should not become harder simply because someone is elderly, alone, or not comfortable with technology. OUR SOLUTION One voice assistant for everyday care + emergencies. Sahara lets patients communicate naturally in Hindi, Hinglish, English, and more languages, with multilingual voice support powered by Murf AI TTS. There is: No typing.No complicated menus.No switching between apps. The entire experience revolves around the patient, with Sahara and its tool-calling ecosystem working around them. WHAT SAHARA DOES 🗣️ 1. Natural Voice Conversations Sahara talks like a person. Real-time two-way voice Patients can interrupt the assistant Conversation context is remembered Previous information can be used even in a later call Supports Hindi, Hinglish, English, and more The patient simply speaks instead of navigating an app. 🛡️ 2. Type-Safe AI Health Decisions with Laya Health decisions cannot depend on unpredictable free-form AI responses. Sahara uses Laya for type-safe AI decisions. Every health decision produces: A fixed, checked response A confidence score A structured output This means a wrong AI-generated sentence cannot directly trigger a critical action. ❤️ 3. Hospital-Standard Risk Scoring Sahara evaluates: Blood pressure Pulse Oxygen saturation Blood sugar Symptoms The system uses NEWS2 (National Early Warning Score 2) principles to classify risk. Risk Levels 🟢 GREEN — Stable🟡 YELLOW — Needs attention🟠 ORANGE — High risk🔴 RED — Emergency A single dangerous reading can be enough to trigger a RED emergency state. 🚨 4. Emergency Alert Chain When the system detects a RED emergency: The caregiver and next contact are alerted. If nobody responds, Sahara continues trying the next contacts. The system retries every 30 seconds. Backup emergency alerts are triggered if nobody responds. A large one-tap 108 button remains available on the patient's screen. The goal is simple: Don't let one missed call become a missed rescue. 🏥 5. Doctor-Style Emergency Handoff When an emergency happens, responders and family shouldn't have to reconstruct the situation themselves. Sahara automatically creates a concise SBAR-style handoff summary containing: What happened Relevant medical history Current risk What should happen next This gives caregivers and responders immediate context. 👁️ 6. "Are You Awake?" Emergency Check During an active emergency, Sahara checks on the patient every 60 seconds. It asks: "Are you awake?" If the patient becomes unresponsive or stops responding: → Sahara automatically sends an urgent WhatsApp alert. 👨‍👩‍👧 7. Family & Caregiver Connection Sahara connects the patient with their trusted care network. Caregiver features Caregiver pairing Live caregiver emergency screen WhatsApp emergency alerts Unresponsive alerts Emergency-resolved notifications Daily medicine reports Care history Vitals trends Family members can stay informed without constantly calling the patient. 💊 8. Everyday Medicine Assistance Sahara isn't only for emergencies. Patients can use their voice to: Ask about today's medicines Say "I took it" to log a medicine Set reminders such as "Remind me at 8 PM" Save BP readings Save sugar readings Track their health history 💰 9. Cheaper Medicine Discovery Branded medicines can be expensive. Sahara helps patients compare medicines with cheaper Jan Aushadhi / government generic alternatives. It currently covers 43 medicines, with live lookup available for additional medicines. The goal is to help patients understand: "Is there a cheaper generic alternative to what I'm buying?" 🇮🇳 10. Government Healthcare Schemes Sahara explains government healthcare schemes in simple language, including: Ayushman Bharat It explains the scheme and its coverage, including treatment benefits of up to ₹5 lakh where applicable. It also covers 12 additional government health schemes, explained in simple Hindi. 📍 11. Nearby Healthcare Discovery Through voice, users can ask Sahara to find: Nearby clinics Medicine stores Healthcare facilities No typing or map navigation is required. 👀 12. Big Visual Health Cards Every voice response can also appear as a large, high-contrast visual card. This means the patient can: Hear it + See it The UI is designed around accessibility rather than information density. 👴 BUILT FOR SENIORS Sahara is intentionally designed for elderly users. Accessibility-first design Hindi / English language switch Three-step onboarding Extra-large touch buttons High-contrast interface Clear Devanagari text Large visual cards Simple interactions "I'm OK" button when speaking is difficult The interface tries to remove the need to understand how an app works. ⚙️ HOW AGORA POWERS SAHARA Agora Conversational AI powers the complete voice conversation. The pipeline includes: Speech → AI → Voice with: Deepgram for speech-to-text AI reasoning Murf for Hindi text-to-speech Agora's real-time audio provides low-delay voice communication, while live captions flow through Agora's data channel. 🧠 OUR AI ARCHITECTURE Our own Python FastAPI server plugs into Agora as the AI brain. Whenever the patient speaks: 1. Patient speaks ↓ 2. Server receives the request ↓ 3. Safety checks run first ↓ 4. Appropriate tool is selected ↓ 5. Health / emergency action is executed ↓ 6. Visual card is generated ↓ 7. AI generates the conversational response ↓ 8. Response is spoken back to the patient This architecture keeps safety-critical actions outside unrestricted AI generation. 📱 PLATFORM SUPPORT Sahara works across: Android iOS Web 🧰 TECH STACK Layer Technology Mobile App React Native + TypeScript Backend Python + FastAPI Voice / RTC Agora Speech-to-Text Deepgram Text-to-Speech Murf AI AI Models Groq + OpenAI + Claude Emergency Communication WhatsApp Business API Maps / Places OpenStreetMap Deployment Render 🔥 THE HARD PARTS WE SOLVED Building Sahara in just 3 days meant solving several difficult problems simultaneously. 1. Emergency reliability + conversational AI The AI needed to feel friendly and natural while emergency actions had to remain predictable and controlled. 2. Hindi & Hinglish understanding Real users don't always speak textbook Hindi or English. Sahara needed to understand natural Hindi + English mixing. 3. Voice + UI synchronization The spoken response, live captions, health data, and visual cards all needed to stay synchronized. 4. Live failure recovery The system needed fallback AI providers and alternate voice flows so the demo could continue even when individual services failed. 5. Duplicate emergency prevention The same emergency must never result in multiple duplicate alert chains. ⚠️ HONEST LIMITATIONS Sahara is a prototype built during a hackathon, and we want to be transparent about what is and isn't connected today. 🚑 Emergency services Ambulance and hospital alerts are designed into the system, but they are not yet connected to real 108/112 or hospital systems. Today: Real alerts go to family members through WhatsApp. The patient has a one-tap 108 call button. 💾 Data storage Data currently lives in memory. A production system would require: Secure database User authentication Proper patient identity management Long-term health record storage 📷 Prescription reading Prescription photo reading is currently being connected to a vision AI. Manual entry works today. 🩺 Medical disclaimer Sahara is a support tool, not a doctor. It is designed to assist patients and caregivers, not replace qualified medical professionals. 🚀 WHAT'S NEXT 🏥 1. Clinic & Hospital Admin Dashboard The next step is taking Sahara beyond the patient app. Healthcare staff could manage follow-ups through a central dashboard. For example: Patient discharged↓Sahara automatically calls them↓"How are you feeling today?"↓Patient responds naturally↓Sahara saves a structured summary↓Only high-risk patients are escalated to a human The same system could automatically follow up after: Hospital discharge Missed medication Risky vitals Missed appointments This can save staff time while making patients feel continuously cared for. 🚑 2. Real Emergency Integration Future versions can integrate with: 108 / 112 Hospitals Ambulance systems Hospital pre-alert workflows So responders can receive the patient's context before they arrive. 🦾 3. Fall Detection Integrate device and phone sensors to automatically detect potential falls and start the emergency workflow. 📡 4. Connected Medical Devices Sahara can connect directly with: Blood pressure monitors Pulse oximeters Glucose meters Other home health devices This removes the need for manual readings. 🗣️ 5. More Indian Languages Expand beyond Hindi and English into: Tamil Bengali Marathi Telugu And more Indian languages All within the same patient-first system. 🪪 6. ABHA & Healthcare Ecosystem Integration Future integrations include: ABHA health records Family mode ASHA worker mode Privacy and consent management 🌱 THE BIGGER VISION Sahara isn't trying to build another healthcare app. It is trying to build a voice-first healthcare layer around the patient. On normal days: "Did I take my medicine?" On difficult days: "What's my BP?" And when something goes seriously wrong: "I can't do this myself." That's when Sahara stops being just an assistant— and becomes a lifeline. Built for the Agora Voice AI Hackathon 2026   Built in 3 days.Designed around one simple idea: Healthcare should listen.