Anusha Sundaramurthi

Problem: Job seekers struggle to bypass ATS filters and tailor resumes efficiently.

Solution: Built an AI-powered resume management and job tracking system using semantic similarity and the Claude API.

Key Features: Centralized PDF storage, real-time application tracking, AI-driven actionable suggestions, and ATS scoring.

Outcome: Delivered 95%+ multi-format extraction accuracy, providing real-time feedback for over 1,000 resumes.

Problem: High operational costs and hallucination risks associated with cloud-based LLM APIs for support.

Solution: Architected a fully local, offline RAG pipeline to query e-commerce PDF docs safely.

Key Features: PDF document QA, semantic search using Sentence Transformers, Top-K retrieval (FAISS), and Streamlit Chat UI.

Outcome: Reduced cloud API costs by 100% and drastically minimized hallucinations across 500+ query benchmarks.

Problem: Complex user queries often require routing to specialized generation models (text, images, or video).

Solution: Designed an agent capable of dynamic intent classification to route prompts across three distinct AI modalities seamlessly.

Key Features: Intent routing mechanism, GPU-accelerated backend via Google Colab, and a local FastAPI proxy layer.

Outcome: Centralized interface handling text, image, and video generation endpoints automatically based on natural language.

Problem: Need for high-speed, lightweight text sentiment analysis without relying on heavy external LLM calls.

Solution: Built an NLP classifier using semantic similarity with transformer embeddings to detect 7 distinct emotional states.

Key Features: Uses all-MiniLM-L6-v2 embeddings, modern HTML/JS frontend with smooth animations, lightweight REST architecture.

Outcome: Deployed a highly responsive API achieving sub-100ms real-time inference latency via FastAPI.

Problem: Fragmented developer experience when testing and comparing outputs across multiple AI providers.

Solution: Engineered a unified, multi-provider AI chat interface powered by FastAPI and Streamlit.

Key Features: Docker containerization for instant setup, secure API key environment management, and active session handling.

Outcome: Created a highly extensible, production-ready foundation suitable for integration into complex Agentic or RAG workflows.

Problem: Missing context-aware, integrated AI refactoring capabilities natively within standard lightweight code editors.

Solution: Developed a context-aware coding environment with semantic indexing, direct chat, and MCP Server integrations.

Key Features: "Add to Chat" / "Add to Edit" snippet actions, directory-wide code embedding for search, Model Context Protocol integration.

Outcome: Boosts coding productivity by enabling semantic repository questioning and automated in-editor refactoring.

Problem: Inefficient agricultural water management leading to both crop stress and systemic resource waste.

Solution: Created an IoT-based system that dynamically tracks real-time soil moisture and local weather to recommend watering schedules.

Key Features: Firebase realtime database integration, OpenWeather API pulls, live dashboard, crop-specific logic, MIT App Inventor UI.

Outcome: Delivered a fully tested cloud and software layer, primed for direct Raspberry Pi hardware sensor deployment.

Problem: Students lacked a dedicated, university-specific tool to accurately model and project their 8-semester grading trajectory.

Solution: Designed a modern, highly responsive frontend utility engineered specifically for the CSE curriculum.

Key Features: Precise credit weighting logic mapped to all 8 semesters, instant aggregate tracking, and mobile-first UX layout.

Outcome: Streamlined complex academic tracking and grade classification into an intuitive, zero-latency static application.