Hi, I'm Sanchit 👋

A job-application workspace that finds postings, scores them against your profile, and prepares tailored resumes and cover letters. I built the Python/FastAPI engine and Next.js interface, with per-user data isolation, encrypted provider keys, a tracker, interview preparation, and a Chrome autofill extension. Bring a cloud model or connect local Ollama. The guided demo uses a fictional candidate and simulated AI responses.

A consumer-trust auditing tool, evolved from DarkGuard. A headless browser scans pages and multi-step flows for manipulative interfaces, then pairs findings with a trust grade, annotated screenshots, recordings, and downloadable reports. I built deterministic and AI-assisted analysis, a dashboard, and a Chrome extension for checking the page you are browsing. Regulatory references help explain each finding and its possible implications.

A private daily journal connecting what you did with how you felt. I built an Expo/React Native app with a 24-hour activity timeline, focus check-ins, small changes to try, and weekly review. SQLite keeps the core journal on-device and available offline without an account. Insights show observation coverage and associations in your logs; missing answers stay unknown rather than becoming zeroes.

Four games for two people spending time together or apart: Doodle Duo, Know Me Better, In Sync, and Clue Quest. I built synchronized play, hidden-answer reveals, reconnect handling, and a shared scorecard using React, Fastify, and Socket.IO. A Three.js cat brings the table to life. The project is under active development; public launch is on hold.

An installable packing tracker that organizes trips into bags, cubes, and items, with checklists and packing progress. I built reactive IndexedDB storage, reusable templates, drag-and-drop organization, and offline support. The core tracker stores data in your browser and needs no account or AI connection. Optional AI tools send the requested trip or packing context to a configured provider to draft lists and suggest forgotten items.

UW independent-study work on translating English into ASL gloss and grammatical non-manual markers. I worked on a Stage 1 reimplementation and evaluation pipeline comparing fixed examples, FAISS retrieval, paragraph context, and their combination. The work examines data leakage, gloss conventions, and qualitative errors alongside automatic metrics. It is translation research, with no claim of a validated sign-language interpreter or complete video synthesis.

An accessibility prototype exploring ASL overlays for YouTube. Earlier work chained retrieved signing clips; the current direction uses an LLM interpretation plan and retrieved motion primitives to drive a 3D avatar, including grammatical non-manual markers. I built the extension and Python pipeline foundations. The avatar system is still in build-out and has not been validated as a replacement for a human interpreter.

A self-hosted platform for storage, backups, virtual machines, and personal web applications. I built on TrueNAS and ZFS, then added container hosting, HTTPS routing through Cloudflare and Traefik, and GitHub Actions image builds. It now supports products such as Applination and Luggist. The work spans service deployment, persistent storage, monitoring, and operating applications beyond a local development machine.

A creativity experiment that takes a topic and a wildness setting, generates several ideas with Gemini, and uses Mistral as an independent comparative judge. I built the best-of-N pipeline, streamed progress, and retro web interface. The judge compares coherence, novelty, and surprise; an explicitly labeled heuristic fallback keeps provider failures visible. The scores guide selection rather than proving an idea has never been thought before.

Full-stack fantasy cricket platform serving 1,000+ users with real-time leaderboards, OAuth2, and Supabase Row-Level Security. Features context-aware AI match summaries using Gemini API with Google Search grounding, adapting to pre/live/post-match states. Currently supports IPL, with extensibility for other leagues.

Real-time mask detection for edge devices (Raspberry Pi), using Haar Cascade for fast face detection with low latency.

Hacker News' Ask HN column in a Newspaper-style UI using ReactJS, NodeJS, and Hacker News API.

Developed a neural network model using PSPNet and TensorFlow to segment sky pixels in images for enhanced outdoor image analysis.

Spielen is an Android app that helps users find, host, and join local sports events with features like map-based location, event reminders, and direct contact with hosts.

Developed a neural network model in TensorFlow and Keras for classifying a given set of parameters into stars, galaxies, and quasars, with a Tkinter GUI

A simple Snake game implemented in C++ using 'time' and 'windows' libraries