Aagnya Mistry

Problem: Manual compliance review across large PDF documents is slow, inconsistent, and hard to scale for enterprise workflows.

Solution: Built a full-stack compliance platform with three specialized AI agents for PDF ingestion and embeddings, iterative RAG reasoning via Groq, and deterministic risk scoring with Excel report export.

Impact: Accelerated compliance analysis with semantic retrieval, configurable loop control, and stakeholder-ready risk reports from a single workflow.

Problem: Task managers usually miss unstructured input and fail to account for burnout risk, leaving users with scattered work and poor scheduling decisions.

Solution: Developed a full-stack Flutter and FastAPI app that captures text, voice, and image inputs, structures tasks with AI and OCR, and generates adaptive schedules using burnout-aware scoring.

Impact: Enabled reliable management of 100+ tasks with high structuring accuracy, real-time sync, and resilient fallback systems for AI, auth, and data services.

Problem: Research papers and news articles are often dense and difficult to understand quickly.

Solution: Developed a RAG-powered assistant that processes PDFs and URLs, enabling users to ask contextual questions using LangChain and Gemini.

Impact: Helped simplify complex research content and automate explanation of articles shared by real users.

Problem: Documenting GitHub repositories for interviews is time-consuming and inconsistent.

Solution: Created an agentic system that scans a GitHub repository and generates a structured Notion-style document summarizing architecture, features, and dependencies.

Impact: Reduced manual documentation time by 90% and improved interview preparation efficiency.

Problem: Educational YouTube and NPTEL lectures are long and dense, making it difficult to extract key insights efficiently.

Solution: Built a RAG-based system using Grok LLM to generate structured notes from video content. Implemented a fallback pipeline using Grok Whisper for audio transcription when YouTube subtitles are missing or unavailable.

Impact: Reduced hours of manual note-taking and enabled faster content revision for students consuming lengthy technical lectures.

Problem: Farmers often struggle to diagnose crop diseases early due to limited access to technical expertise.

Solution: Developed a real-time mobile app using Flutter and TensorFlow/OpenCV to detect plant diseases and provide preventive advice.

Impact: Selected as a Smart India Hackathon 2024 finalist; improved disease identification accuracy for multiple crop categories.

Problem: Recidivism prediction models can unintentionally discriminate across sensitive demographic groups.

Solution: Conducted bias detection and mitigation using SHAP and FairLearn to improve fairness metrics without major accuracy loss.

Impact: Enhanced model interpretability and reduced bias, demonstrating responsible and ethical AI practices.

Problem: Traditional payment tracking systems lack transparency and real-time verification.

Solution: Built a distributed-ledger-inspired system using Spring Boot and TypeScript, deployed using Docker for isolated microservices.

Impact: Successfully deployed prototype used for 1000+ test transactions, demonstrating backend + Docker deployment skills.