HaradeepPuneti

Real mock interviews are scarce, expensive, and inconsistent, and most prep tools are text chat. They miss what a system-design or DSA interview actually is: spoken, whiteboard-driven, and graded on judgment under time.

A voice-first AI interviewer that runs phase-gated system design, OOD, and algorithm sessions over a shared whiteboard and code pane. It streams a real-interviewer-styled Claude response, speaks it aloud, watches what you draw, and ends with a deterministic, transcript-grounded scored report.

• You talk, it listens, and it talks back. Your speech is transcribed, the AI interviewer replies in real time, and its answer is spoken aloud, so it feels like a live conversation rather than a chat box.
• It actually reads your whiteboard. Whatever you sketch is turned into something the AI can see, so it asks about your diagram instead of ignoring it.
• Two AI models share the work: a stronger one runs the interview and grades you, a faster, cheaper one handles quick things like hints, which keeps replies snappy and the cost down.
• Your code runs for real in a locked-down sandbox, and the final score blends how many hidden tests passed with the AI's judgment, so it rewards working code, not just good-sounding answers.

• 242 questions across system design, object-oriented design, and algorithms, with a free first interview and pay-as-you-go after that.
• A caching trick cut the cost of each conversation turn to about a tenth of what it would normally be, which keeps each interview cheap to run.
• Scoring is built to be fair and hard to game: the same answers always earn the same grade, user code is isolated so it can't break anything, and faked scores can't sneak onto the leaderboard.
• Backed by a large automated test suite and a nightly check that has the AI grade sample interviews, so quality does not quietly slip.

A family farm selling at the Blacksburg farmers market needed an online store that matched how they really sell, by season and by booth pickup, without forcing customers to make accounts, plus a back office to run inventory and orders.

A single-vendor storefront with no-account guest checkout and an admin dashboard, with payments and physical-booth inventory both synced through Clover so the website and the in-person register never drift apart.

• Customers can browse and check out as guests, with no account required, and only ever see what's actually in season.
• The website and the farm's in-person card reader stay in sync: a sale at the market updates the site's stock automatically, and updating the site updates the reader.
• Search understands plain language instead of exact keywords, so 'tomatoes' still turns up 'heirloom tomato.'
• The farmer can snap one photo of a product and the site drafts the listing, name, description, and price, for them to tweak and publish.

• Orders are handled exactly once, even when the payment system sends the same notification twice, so customers are never double-charged and stock is never double-counted.
• Sold-out or removed products still show up correctly on past receipts and order history.
• Automatic emails at every step of an order, plus an admin dashboard with revenue, a sales chart, and a harvest calendar the owners actually use.

Built at the VT Google Developer Group hackathon by a team of three and taken to production. I was the top contributor (~76 of 141 commits), leading the checkout and order pipeline, the email system, analytics, and the migration from Stripe to Clover.

SAT and PSAT prep tools make you hunt through static problem sets by topic. Students can't just say what they want to work on, like "comma usage" or "quadratic equations," and get relevant real-exam questions back.

A full-stack prep app built around natural-language search over a curated bank of real-format questions, with per-subject progress, spaced-repetition review, and a Pro tier, covering old SAT, Digital SAT, Digital PSAT, and SAT Math.

• Type what you want to practice in plain English, like 'comma rules' or 'systems of equations,' and it pulls matching real-exam questions, skipping ones you've already done.
• Questions you miss come back on a spaced-repetition schedule, timed to resurface right before you'd forget them.
• Answers are checked on the server, not in the browser, so the questions and their answers can't be peeked at or gamed.
• Behind the scenes, an AI pipeline reads raw exam questions and tags them by skill, which is what makes the plain-English search accurate.

• Around 1,000 curated questions across four formats: old SAT, Digital SAT, Digital PSAT, and SAT Math.
• Free to try with a few practice sessions a day, plus a paid tier for unlimited practice and step-by-step explanations.
• Rebuilt the search engine onto a faster, cheaper setup without having to reprocess the entire question bank.