N.O.V.A.
An AI interview preparation and learning system built on a multi-agent architecture that uses a six-step teaching loop—question generation, evaluation, follow-up questioning, explanation, weakness tracking, and personalized planning—to help users learn from mistakes rather than simply drilling questions.
Timeline (last 90 days)
N.O.V.A. is an AI interview preparation tool built on Next.js and NVIDIA model serving, using a multi-agent architecture with a six-step learning loop.
N.O.V.A.'s flow has six stages: generating questions by candidate level, evaluating on four dimensions of accuracy/clarity/completeness/reasoning, adaptive follow-up questions, explaining concepts when stuck, tracking weaknesses and misconceptions, and generating a personalized preparation plan.
When a candidate answers "I don't know", the system records the attempt as 0 points, and a teacher agent outputs a structured lesson containing a plain-language explanation, technical details, real-world examples, optional diagrams, a concise answer usable in interviews, common mistakes, and follow-up questions.
N.O.V.A. stores wrong answers in an "error notebook" and reviews them at intervals of 1→2→4→7→15→30 days.
N.O.V.A. uses SQLite as its database, using the built-in node:sqlite locally and a Turso embedded replica in the cloud to avoid data loss on hosts with ephemeral disks.
N.O.V.A. calls models through NVIDIA's OpenAI-compatible API, with keys stored encrypted and never sent to the browser.
N.O.V.A. uses a multi-agent design in which each role (interviewer, question generator, answer evaluator, teacher, coding interviewer, code reviewer, system design evaluator, hiring committee, etc.) is its own agent, and each agent returns strict JSON output.
Reasoning models sometimes take 20–70 seconds to respond and may truncate JSON.
When the first question takes more than 12 seconds or JSON is invalid, N.O.V.A. falls back to a human-verified question bank and scores by keyword coverage, and the report notes the source.
N.O.V.A. runs end to end even with AI turned off.
3 more timeline events
All Facts (13)
N.O.V.A. is an AI interview preparation tool built on Next.js and NVIDIA model serving, using a multi-agent architecture with a six-step learning loop.
50%UnverifiedN.O.V.A.'s flow has six stages: generating questions by candidate level, evaluating on four dimensions of accuracy/clarity/completeness/reasoning, adaptive follow-up questions, explaining concepts when stuck, tracking weaknesses and misconceptions, and generating a personalized preparation plan.
50%UnverifiedWhen a candidate answers "I don't know", the system records the attempt as 0 points, and a teacher agent outputs a structured lesson containing a plain-language explanation, technical details, real-world examples, optional diagrams, a concise answer usable in interviews, common mistakes, and follow-up questions.
50%UnverifiedN.O.V.A. stores wrong answers in an "error notebook" and reviews them at intervals of 1→2→4→7→15→30 days.
50%UnverifiedN.O.V.A. uses SQLite as its database, using the built-in node:sqlite locally and a Turso embedded replica in the cloud to avoid data loss on hosts with ephemeral disks.
50%UnverifiedN.O.V.A. calls models through NVIDIA's OpenAI-compatible API, with keys stored encrypted and never sent to the browser.
50%UnverifiedN.O.V.A. uses a multi-agent design in which each role (interviewer, question generator, answer evaluator, teacher, coding interviewer, code reviewer, system design evaluator, hiring committee, etc.) is its own agent, and each agent returns strict JSON output.
50%UnverifiedReasoning models sometimes take 20–70 seconds to respond and may truncate JSON.
50%UnverifiedWhen the first question takes more than 12 seconds or JSON is invalid, N.O.V.A. falls back to a human-verified question bank and scores by keyword coverage, and the report notes the source.
50%UnverifiedN.O.V.A. runs end to end even with AI turned off.
50%UnverifiedN.O.V.A. runs Python with a time limit via Pyodide in a Web Worker, runs SQL via SQLite compiled to WebAssembly, and keeps reference solutions on the server until the user solves the problem.
50%UnverifiedN.O.V.A. updates mastery using an exponential moving average (EMA), with the formula new mastery = α × this score + (1-α) × old mastery, where α is typically 0.2–0.4.
50%UnverifiedThe author admits there is currently no rigorous evaluation method to prove the effectiveness of the feedback, judging only by whether mastery rises after the "I don't know" lesson plus a subsequent retry.
50%