Bhanu Sanikommu

Full-Stack Developer

Saveetha School of EngineeringChennaiinternship, freelance

Open to roles

LeadershipGITProblem SolvingFrontend Developer

Personal website· my-portfolio-psi-orcin-70.vercel.app GitHub LinkedIn

Chase the Cloud Challenge '26 (ISRO)

SkyMind AI is an AI-powered cloud motion prediction and weather intelligence platform built for the ISRO Chase the Cloud Challenge 2026. The system predicts future cloud movement using a modular AI nowcasting pipeline and provides an interactive mobile-first weather intelligence dashboard. Key Features • AI-powered cloud motion prediction pipeline • Multi-channel satellite image preprocessing pipeline • Diffusion model architecture with baseline forecasting models • Optical-flow assisted prediction pipeline • Real-time weather dashboard • Interactive radar and satellite visualization • AI Weather Assistant powered by Google Gemini with Groq fallback • Explainable AI (XAI) visualizations • Prediction confidence estimation • Automatic PDF, CSV and JSON report generation • REST API with OpenAPI documentation • Mobile-first Progressive Web App • Offline-ready architecture • Docker support • Comprehensive documentation Project Structure • Next.js + React frontend • FastAPI backend • Modular ML pipeline • Evaluation framework • Report generation engine • AI orchestration layer Evaluation The project includes automated evaluation using SSIM, PSNR, MAE, RMSE, IoU and additional computer vision metrics. Current Status The architecture, preprocessing pipeline, APIs, evaluation framework, reporting system and AI assistant are fully implemented. The project is designed to train directly on official INSAT/MOSDAC satellite imagery. Access to official MOSDAC datasets requires user credentials. The ingestion and training pipeline is included and documented, but demonstration training is currently limited by dataset access. Repository includes: • Complete source code • Documentation • Setup guide • Architecture diagrams • API documentation • Deployment instructions Thank you for reviewing SkyMind AI.

5 media files · github.comView

AI-Powered HR Operations & Employee Experience Challenge

Video url-https://drive.google.com/file/d/1lbFaxu3pXi48pJk0v5g6aD-Y_62YdHTD/view?usp=drivesdk Live demo-https://pulsehr-ai-tawny.vercel.app/ GitHub repo:https://github.com/sanikommu-bhanu/pulsehr-ai 📱 WHAT THIS APP DOES PulseHR AI is one HR platform with two connected apps sharing live company data: an Employee app and an HR/Admin app. Sign up as HR to create a company (get a 6-char join code); sign up as Employee using that code to join it. From there, every shared action updates both sides in real time — no refresh, no polling delay. Modules (all real, working CRUD — not mockups): • Leave — apply, cancel, HR approve/reject, live balance computed from actual approved requests, team calendar • Attendance — check in/out, history, monthly summary • Helpdesk — raise a ticket, AI pre-chat triage, HR moves it Open → In Progress → Closed, employee notified instantly • Company Feed — HR posts announcements, everyone sees them live (employees can't post — enforced by database rules, not just UI) • Notifications — real-time feed from actual events: leave decisions, ticket updates, new payslips, new feedback • Payroll — HR issues a payslip (gross/deductions → net), employee sees it appear instantly • Performance — peer-to-peer feedback (real, live), not a fabricated star rating • AI Assistant — Gemini-powered chat that reads live company data (leave, attendance, tickets) and can autonomously file a helpdesk ticket from a plain-English request like "my laptop is broken" • Employee directory — live company roster, searchable by both roles 🎯 PROBLEM HR teams juggle leave approvals, attendance, onboarding, and helpdesk tickets across disconnected tools. Employees get no visibility into their own requests, so they follow up by email/Slack, and HR spends time on routine status questions instead of real work. ⚙️ HOW THE SOLUTION WORKS Every shared record (leave, attendance, tickets, payroll, feedback) is written with a real companyId + employeeId. Both the employee and their HR/Admin subscribe to the same data live, so an HR approval or an employee submission is visible on the other side within seconds — demonstrated in two side-by-side tabs (see HOW TO TEST above). Security is enforced at the database level (Firestore rules / a local-mode equivalent), so one company can never read another's data — this isn't just hidden in the UI. The AI Assistant is grounded in this same live data: it never invents a number it can't actually see, and can take real actions (like filing a ticket) rather than just chatting. 📈 EXPECTED BUSINESS IMPACT • Fewer manual status-check interruptions for HR (leave/ticket state is visible to employees in real time) • Faster leave/ticket turnaround since approvals push instantly instead of waiting on email or portal refreshes • One dashboard instead of five disconnected tools for HR to monitor attendance, leave load, and open tickets • AI-assisted ticket triage reduces manual intake work 🔍 SCOPE — BUILT VS. NOT Real: auth, company/employee model, leave, attendance, helpdesk, announcements, notifications, payroll issuance, peer feedback, AI assistant, all enforced by real security rules. Not built (explicitly labeled "Coming soon" in-app, not faked): QR/biometric check-in, OCR/resume parsing, real payroll tax processing, transactional email/SMS/push delivery — each needs a paid provider or hardware API beyond a frontend build.

5 media files · pulsehr-ai-tawny.vercel.appView

MemoryVerse AI '26

PDF link -https://drive.google.com/file/d/1ifU0GKLQjhwCGXM8IEjbjkuSfUevVFrE/view?usp=drivesdk The pdf consists of all the details about project including architecture diagram and Everything in the application is fully functional and properly integrated, except for the LinkedIn integration, which will be implemented in a future update. The other required documents -architecture and though_process are available in my GitHub inside docs folder MemoryVerse AI is an AI-powered Digital Identity platform designed to intelligently organize, understand, and retrieve professional documents such as resumes, certificates, project reports, and internship records. The system leverages Cohere Command-R-08-2024 for document analysis, Embed-English-v3.0 for semantic vector generation, Retrieval-Augmented Generation (RAG) for grounded AI responses, and Semantic Search using the Cosine Similarity algorithm. Built with Next.js 14, React, Firebase Authentication, Cloud Firestore, and Firebase Storage, the platform provides secure, scalable, and intelligent document management. Its key features include AI-powered document categorization, knowledge graph visualization, professional journey timeline, AI chat assistant, resume builder, and portfolio generator, offering an innovative solution that transforms static portfolios into dynamic, searchable digital identities.

5 media files · memoyverse-ai.vercel.appView

QueueCare — Real-Time Clinic Queue Management with Atomic Token Locking

Indian clinics still use paper chit tokens. Patients arrive, get a slip, and sit for hours with zero visibility — no ETA, no position updates, no idea when to return. Receptionists manually shout token numbers, causing missed calls and chaos. Worst of all, when two receptionists manage the same queue, they can accidentally call the same patient — the "double-token" race condition. We identified 3 core pains: (1) patient anxiety from zero wait visibility, (2) receptionist overload from manual queue ops, and (3) data concurrency bugs in existing digital queue tools. [24/06, 2:33 pm] anu: Indian clinics still use paper chit tokens. Patients arrive, get a slip, and sit for hours with zero visibility — no ETA, no position updates, no idea when to return. Receptionists manually shout token numbers, causing missed calls and chaos. Worst of all, when two receptionists manage the same queue, they can accidentally call the same patient — the "double-token" race condition. We identified 3 core pains: (1) patient anxiety from zero wait visibility, (2) receptionist overload from manual queue ops, and (3) data concurrency bugs in existing digital queue tools. [24/06, 2:33 pm] anu: We started by visiting 3 local clinics in Bangalore and observed the full patient journey — from walking in to walking out. We mapped the pain points into a Mermaid sequence diagram and Built a production-ready system with: (1) Zero race conditions — PostgreSQL atomic locking guarantees no two receptionists can call the same token, even at the exact same millisecond. (2) Real-time sync — Socket.io broadcasts update all screens in <200ms. (3) Patient self-service — QR scan → join queue → live tracking with voice announcements. (4) Resilient state — Zustand persists tokens to localStorage so patients don't lose their spot on tab close. (5) Dynamic ETAs — receptionists adjust avg consultation time mid-day, instantly updating all patient screens. Three things: (1) I'd add automated load testing using k6 or Artillery to stress-test the FOR UPDATE SKIP LOCKED query under 500+ concurrent "Call Next" requests — we proved correctness logically but didn't benchmark at scale. (2) I'd implement a Redis pub/sub layer instead of in-memory Socket.io for horizontal scaling across multiple serverless instances. (3) I'd add an SMS fallback for patients without smartphones using Twilio, since not all clinic patients in India have reliable internet. These are honest gaps I'd close in a v2.

20 media files · queue-care-chi.vercel.appView

This is Bhanu’s work on Wooble.

Build a profile that shows what you can do — and share it anywhere.

Build yours