Pokédex

Problem Solved

Traditional learning platforms force every student through the same curriculum regardless of their pace or understanding. LearnFlow AI continuously evaluates mastery, generates personalized study paths, and uses Socratic questioning to guide students toward deeper conceptual understanding rather than simply providing answers.

Problem Solved

Helping homeless individuals often requires coordination between multiple organizations, resulting in delayed responses and duplicated efforts. Aasrah creates a unified workflow where citizens submit reports, NGOs manage cases, volunteers choose or receive assignments, and every report remains traceable throughout its lifecycle.

Problem Solved

Existing OCR systems perform poorly on degraded historical Sanskrit manuscripts due to image noise, faded ink, and handwritten scripts. DivyaLipi AI combines document localization, OCR, image preprocessing, and LLM-assisted translation to recover readable digital text from centuries-old manuscripts.

Problem Solved

Managing hundreds of megabytes of event photos on free hosting while allowing non-technical administrators to publish new events is difficult. The platform uses GitHub as a headless CMS, jsDelivr as a global CDN, and runtime event discovery to deliver scalable media hosting without requiring redeployment or paid infrastructure.

Problem Solved

Developers often rely on multiple disconnected tools for repository analysis, vulnerability detection, and code review. Qwerty AI Scanner consolidates these workflows into a unified interface that analyzes projects, identifies potential issues, and provides AI-generated explanations and actionable recommendations.

Problem Solved

Traditional portfolio websites are static and forgettable. This project recreates the feel of Pokémon FireRed in the browser with a custom tile engine, collision detection, layered maps, NPC interactions, building interiors, camera movement, and interactive project exploration to create a memorable developer portfolio.

Problem Solved

A small farm stay needed a modern, indexable website without a CMS or backend. The site keeps all copy as typed content, paints dawn-to-night palettes before hydration, and ships structured data so search and social previews stay accurate while bookings stay on Airbnb.

Problem Solved

Client-rendered SPAs leave search engines and WhatsApp scrapers with an empty shell. This site uses a Vite SSR + prerender pipeline so every route ships real HTML, metadata, and JSON-LD while staying easy to maintain in plain React.

Problem Solved

Homelab Minecraft ops were split across fragile scripts for Discord bridging and idle shutdown. One typed, restart-safe daemon now streams logs, tracks players, auto-stops idle servers, and keeps the whole path covered by a large automated test suite.

Problem Solved

Creators often bounce between chatbots and docs to go from idea to shootable script. Drafter stages the whole pipeline — refine, angle, hooks, research, scene script, and visual directions — as a regeneratable workspace instead of a one-shot chat.

Problem Solved

Generic chatbots invent answers about uploaded documents. This pipeline chunks PDFs with page metadata, embeds into ChromaDB, and generates answers with citations so every claim stays tied to source pages.

Problem Solved

Keystroke-path search needs single-digit millisecond lookups, relevance under both all-time and trending scores, and batched write pressure. This service implements prefix ranking, an in-process distributed cache, and aggregated counter updates against the ORCAS dataset.

Problem Solved

Long ChatGPT/Claude/Gemini threads lag because every message stays in the DOM. LagClear applies content-visibility virtualization to off-screen turns so a 2,000-message thread paints about as cheaply as a short one.

Problem Solved

Agent runs need a reliable execution and evaluation harness around Claude Code. Waltuh separates decision from execution and ships doctor checks, acceptance tests, and published benchmarks so agent behavior can be verified instead of trusted blindly.

Problem Solved

Developers struggle to see how recruiters read their GitHub. The analyzer scores profiles, reviews repository health, and generates README-oriented recommendations in a single responsive UI.

Problem Solved

Natural-language dates are messy and inconsistent. This project trains an attention seq2seq model to map informal date phrases into structured outputs and visualizes where the model attends while decoding.

Problem Solved

Generic tutoring chatbots lack a consistent teaching voice. Separate mentor personas (DSA, career strategy, intuitive explanations) keep tone and guidance style stable while sharing one LLM backend.

Problem Solved

Training results are hard to communicate without interactive evidence. The showcase surfaces PR/F1 curves and prediction exploration so model quality is visible without digging through training logs.

Problem Solved

A single alert does not tell you whether Claude finished, needs permission, or is waiting on a question. These plugins play distinct tones when the terminal is unfocused and add wall-clock turn timelines for past sessions.