Where developers grow together

Interview ai agent

Interview AI Assistant is a voice-native adaptive AI interview panel built for the Agora EchoSphere 2026 Hackathon.   Most AI interview bots ask a fixed list of questions regardless of how the candidate answers. This one adapts — a Interviewer Agent conducts the live conversation while a separate Evaluator Agent silently scores each answer for clarity, relevance, depth, and confidence in real time. That score decides whether the next question goes deeper, moves to a new topic, or offers a supportive follow-up — so no two interviews play out the same way.   Built with: - Lyzr Agent Studio for multi-agent orchestration (Interviewer + Evaluator agents working from shared state) - Agora Conversational AI Engine + RTC SDK for real-time voice, turn-taking, and interruption handling - Structured JSON scoring pipeline feeding a live scorecard and a final evaluation report   Why I built it: early-stage interview screening doesn't scale well for recruiters, and most "voice AI" demos are just chatbots with a microphone bolted on — they don't actually reason about what they hear. This project explores what a genuinely adaptive, multi-agent voice interview looks like, and could extend into real hiring screening, mock-interview practice for candidates, or training tools for junior interviewers.   Team Ultron — Adarsh Sarode & Jigar Kumeriya

SentinelAI — Real-Time AI Incident Commander

SentinelAI is a real-time conversational AI incident commander designed to help technical teams coordinate during live operational incidents. It joins an Agora-powered voice room as an active AI participant, listens to the conversation, understands participant roles, and continuously builds a structured Live Incident State. SentinelAI separates confirmed facts from hypotheses, tracks decisions, actions and ownership, detects missing or conflicting information, maintains a live incident timeline, and surfaces unresolved risks. It can provide spoken status updates and clarification requests while keeping humans in control of critical actions through explicit approval. The prototype demonstrates how real-time voice communication, conversational AI, and AI-agent intelligence can work together to create faster, safer, and more accountable incident response.

RuleGit

This project tries to present goverment act, rules, policies, SOPs in versioned structure. Each of the modifying notification, circular or even clarifying instruction bumps the version.    As the name suggests, it is git but for rules. It has functionality to compare different versions, see what individual instructions added, deleted, modified which paragraph or section of policy. With this the visitors are able to see the policy rules existing at any point of time