OUMAI

Proactive Multi-Agent Intelligence an autonomous ecosystem.

Comment

Inspiration

The inspiration for OUMAI comes from the dual life led by thousands of students in tech hubs like Bengaluru—first-generation college students who are pursuing B.Tech degrees by day, but remain the primary "tech support" for their farming families by night. We realized that while AI is advancing rapidly, it remains inaccessible to the people who need it most: The Information Gap: Farmers struggle with volatile Mandi prices and crop diseases, often lacking the latest agricultural tech. The Student Burden: Engineering students from rural backgrounds are under immense pressure to excel in their studies while simultaneously helping their parents navigate a digital world full of scams. The Health Crisis: Rising pollution and grueling labor in both the fields and the city are causing unprecedented health issues that often go ignored until they become emergencies. We built OUMAI (Aegis OS) to be a "Digital Bridge." It is designed to be the "Smart Buddy" for the student and a "Protective Shield" for the parent—a single, autonomous ecosystem that turns a student's smartphone into a career architect, a crop doctor, a market analyst, and a scam sentinel all at once. We wanted to move AI away from being a luxury toy and turn it into a survival tool for the modern Indian family.

What it does

OUMAI (Aegis OS) is a proactive, multi-agent ecosystem designed as a "Digital Bodyguard" for the modern Indian family. It transforms a single interface into a high-level command center that handles the four most critical pillars of a student's and a farmer's life: 🚨 1. The Sentinel (Security & Scam Protection) As digital scams rise, OUMAI acts as a first line of defense. It doesn't just scan messages; it analyzes the intent behind URLs and phone numbers. If a student's parent receives a "Bank Account Blocked" SMS, the Sentinel intercepts the link, runs a heuristic risk analysis, and provides a clear Red/Green safety score to prevent financial loss. 🎓 2. The Academy (Universal Growth Engine) For the B.Tech student, OUMAI is a Career Architect. It maps out their entire academic journey from 1st year to placement. It audits resumes against 2026 industry standards and identifies "Skill Gaps," providing a personalized roadmap to ensure the student becomes the high-earning professional their family is counting on. 🌾 3. The Harvest (Agri-Intelligence) OUMAI brings the laboratory to the field. Using Edge-AI, it allows a farmer to upload a photo of a leaf to detect diseases offline. Simultaneously, it pulls Live Bengaluru Mandi Data, providing real-time price trends (Tomato, Onion, Potato) so the family knows exactly when and where to sell for the highest profit. 🏥 4. The Vitality (Proactive Health) To combat the health issues caused by pollution and labor, OUMAI features an AI Triage Engine. It monitors symptoms, offers mental wellness "Pulse Checks," and provides basic medical guidance. Crucially, it includes a "108 Emergency Trigger" that bypasses chat and activates immediate help if critical symptoms are detected. 🔄 5. The Cross-Agent "Domino" Effect What makes OUMAI unique is that these agents talk to each other. Example: If the student is stressed about an upcoming exam (Vitality), the system automatically adjusts their study roadmap (Academy) and checks if the family's crop prices are stable (Harvest) to give the student peace of mind.

How we built it

We built Aegis OS by focusing on three core pillars: High-Security Validation, Multimodal Agent Intelligence, and Premium Visual Transparency.

  1. The Architectural Foundation (Next.js 14 & Framer Motion) We chose Next.js 14 (App Router) for its server-side performance and optimized routing. To give it that "Aegis" feel—premium, futuristic, and alive—we used Framer Motion for state-driven animations, like the pulsing status indicators and the glassmorphism effects on the agent cards.
  2. The Security Core Before any AI agent executes a command, it passes through our Neural Processing Flow. We built a multi-layered gate using Zod for strict input schemas and a custom Intent Gate that detects prompt injection or malicious intent. This ensures that the system remains "Agentic" but safe.
  3. Domain-Specific Intelligence We didn't just build one AI; we built a Collaborative Expert Matrix: Sentinel: Uses GPT-4 Turbo to perform deep heuristic analysis on incoming data. Harvest: Directly connects to regional market APIs with custom Sparkline components for technical analysis. Vitality: Implements clinical PHQ-2/GAD-2 screening logic combined with a local medical heuristic engine that can function even when the backend API is offline. Academy: Features a sophisticated PDF-to-intelligence engine that leverages local storage and AI to provide immediate educational value.
  4. The "Autonomous" Demo Hook To make the "Autonomous" part of the project title feel real during a demo, we built a dedicated System Activity Sidebar. This is an auto-scrolling terminal that logs background agent actions (like "Sentinel intercepted a script") in real-time. This proves the agents are working even when the user isn't interacting with them.
  5. Transparency & Trust To solve the AI hallucination problem, we implemented the Source Citation system. Every time an agent gives sensitive advice, it generates a "Book" icon. Clicking this opens a Vault Preview, showing the exact source document that the AI used to retrieve its knowledge.
  6. Demo Robustness We built a Hybrid Backend. While it fully supports OpenAI's GPT-4, we implemented "Local Heuristic" fallbacks in and AegisHerald. If the internet lags or an API key is missing, the system detects keywords and provides a deterministic expert response, ensuring your presentation never breaks.

Challenges we ran into

Contextual Handovers: Ensuring that the Aegis Sentinel could pass "threat context" to Aegis Vitality (the health agent) without losing the original user intent was difficult. We solved this by implementing a Shared Global State within the LangGraph orchestrator. Browser-Based ML: Running large models in a browser environment is resource-heavy. We had to optimize our WebGPU calls and implement model quantization (4-bit) to ensure the app stays responsive on mid-range devices used by farmers. Multimodal Ingestion: Parsing a mixed input (e.g., a voice note and a PDF resume) required building a custom Sequencer that validates and extracts data asynchronously before the agents begin their work.

Accomplishments that we're proud of

The "Domino" Logic: Successfully building a system where a security alert (Sentinel) triggers a stress-check (Vitality) and a career backup plan (Architect) automatically. Real-World Data Integration: Integrating live Mandi JSON feeds into a visually intuitive dashboard for farmers. User Empowerment: Creating a "Zero-Trust" system that prioritizes user safety and privacy through local inference, which is a rare feature in student-level hackathon projects.

What we learned

We learned that "More agents is not always better." Initial versions were too complex; we had to refine the architecture to focus on Cohesion. We also mastered the art of Agentic Memory, realizing that for a B.Tech student, the AI must remember their 1st-semester performance to give a 4th-semester career tip. Finally, we learned the importance of Human-in-the-loop AI—Aegis doesn't just act; it suggests and explains why.

What's next for OUMAI

The future of OUMAI is going to be crazy,we are gonna implement much more features such as Aegis Voice Pro: Moving from basic Speech-to-Text to Full-Duplex Real-time Voice, allowing farmers to have natural, hands-free conversations with the system in regional languages. Blockchain Traceability: Integrating a ledger into Aegis Harvest so farmers can prove the origin and quality of their crops to get better prices from urban buyers,and we are planning for a soil test option too. Community Vaults: Allowing farming cooperatives or student groups to share "The Vault" resources, creating a collaborative knowledge ecosystem. Hardware Integration: Connecting to low-cost IoT soil sensors and wearable health trackers for real-time autonomous monitoring. We are planning to expand this to all fields so that this doesnot get limited only to farmers and btech students,we want it to guide and uplift all the lower class and middle class families.

Built With

  • framermotion
  • localheuristicengine
  • locallibrary
  • lucidereact
  • mockmandiapi
  • next.js
  • openaigpt-4turbo
  • phq-2&gad-2matrices
  • react
  • tailwind
  • typescript
  • vanillacss
  • zod
Share this project:

Updates