Where developers grow together

Saarthi

Saarthi – Your AI Co-Teacher for the Live Classroom   Saarthi is a real-time AI teaching assistant designed to identify and address students' learning gaps during a live classroom session. It quietly listens to the classroom conversation, understands the ongoing lesson, detects confused or repeated student questions, and speaks only during a natural pause to provide a short and relevant explanation. This allows teachers to focus on teaching while Saarthi supports students who may hesitate to ask questions.   The system is built using Next.js, TypeScript, Tailwind CSS, Agora RTC, and Agora Conversational AI Engine, with Deepgram for Speech-to-Text, OpenAI for the language model, and MiniMax for Text-to-Speech. It also includes live transcripts, learning-gap insights, teacher controls, and post-class summaries. A Demo Mode allows the complete experience to be demonstrated even without live API credentials.

Orbit

Project Title ORBIT — Operational Response and Briefing Intelligence Team Project Description ORBIT is a production-oriented, real-time AI incident commander that helps technical, operational, support, and business teams coordinate during critical incidents. Using Agora Conversational AI as its core voice layer, ORBIT joins a live incident room, recognizes participant roles, supports multilingual conversations, and converts unstructured discussions into a continuously updated shared operational picture. ORBIT classifies statements as confirmed facts, hypotheses, decisions, or action items. It maintains an incident timeline, assigns and follows up on tasks, records evidence provenance and confidence, detects contradictory or missing information, and tracks unresolved risks. Its multi-agent architecture includes Commander, Listener, Evidence, Conflict, Timeline, Action, Investigation, Integration, Prediction, and Recovery agents working over an Incident Digital Twin and Evidence Knowledge Graph. ORBIT integrates with tools such as Jira, Slack, PagerDuty, monitoring systems, runbooks, and historical incident databases. It provides spoken status summaries, role-specific briefings, similar-incident retrieval through RAG, anomaly correlation, severity estimation, action escalation, recovery verification, incident replay, analytics, and automatic postmortem generation. Every critical action—including rollback, failover, production restart, or infrastructure modification—requires explicit human approval. Following feedback from our first mentorship round, ORBIT evolved from a reactive assistant into a predictive operational-intelligence engine. Its Prediction Engine correlates telemetry such as latency, errors, traffic, saturation, deployments, dependencies, and customer reports to estimate incident probability and time-to-impact. Dependency-graph intelligence predicts failure propagation and blast radius, while geospatial intelligence identifies affected regions and expanding customer impact. An Incident Simulator allows teams to compare mitigation strategies—such as rollback, traffic rerouting, failover, and capacity scaling—before execution. ORBIT presents expected outcomes and secondary risks for human confirmation. A learning layer evaluates forecast accuracy, detects model drift, uses historical incident knowledge, and improves future recommendations. Our payment-system outage prototype demonstrates live voice participation, evidence organization, contradiction detection, predictive graphs, regional impact analysis, mitigation simulation, human-controlled execution, recovery verification, and final incident reporting. ORBIT does not claim to independently determine root cause. It organizes evidence, highlights uncertainty, predicts emerging risks, and recommends informed actions—while the human incident commander always makes the final decision. Vision and Impact Our vision is to transform incident response from fragmented, reactive firefighting into predictive, coordinated, explainable, and human-controlled operations. ORBIT can reduce confusion, accelerate detection and recovery, preserve organizational knowledge, minimize customer impact, and help teams make safer decisions during high-pressure incidents.

INCYRA: AI-Powered Real-Time Incident Command Center

INCYRA is an AI-powered real-time incident command center that helps teams collaborate during critical technical incidents through live voice rooms. It listens to conversations, captures and correlates incident facts, detects conflicting information, extracts action items and decisions, maintains a real-time incident timeline and provides contextual AI assistance for faster investigation and mitigation. Built using React, Node.js, Express, JavaScript, Agora RTC for real-time voice communication and an AI processing engine, INCYRA is useful for DevOps, SRE, engineering, IT operations and incident response teams that need faster, more organized coordination during outages and critical incidents.