Inspiration

I'm an international student on an F-1 visa. Every week my friends and I ask the same questions: When do I file for OPT? Does my degree qualify for STEM OPT? How many chances do I get at the H-1B lottery? The answers are spread across USCIS, Study in the States, the Federal Register and a lot of outdated blog posts. And the rules keep changing: a weighted H-1B lottery, a $100,000 fee that was later struck down, newly proposed OPT fees, and a duration-of-status rule that a court blocked. There are 1.18 million international students in the US, 57% of them in STEM, and more than 294,000 working on OPT (IIE Open Doors 2025). A mistake on a date can cost someone their status. Personalized plans exist, but they cost money or come only through a university. I wanted a free tool that reads today's official rules and turns them into one student's own plan.

What it does

Bridgepath is an AI agent that builds a sourced career and visa roadmap for an international student.

  1. Tell it where you are: degree, major, STEM or not, graduation date, OPT dates, the job you want, and optionally your resume.
  2. Pick a goal: the agent reads your background and suggests three realistic goals, such as big tech, a cap-exempt university lab, or a stretch path. You pick one or type your own.
  3. Get your roadmap: ◦ Key dates with countdowns: when to ask your DSO, your OPT filing window, grace period, STEM OPT window and H-1B season, plus an Add to calendar button that sets reminders a week and a day before each date. ◦ Your H-1B chances: how many lottery seasons you get while your work authorization lasts, and how the wage-level weighted lottery changes your odds. ◦ Ordered steps with cost, time and documents, each linked to its official source. ◦ Employers to target, where sponsorship is more likely. ◦ Policy watch: each recent rule change with its real status today (in effect, proposed, blocked by a court or struck down), plus this month's news.
  4. Track and ask: tick off steps, ask follow-up questions, download a PDF, and reopen past plans under My roadmaps, saved in your browser with no account needed. A second mode helps nurses, engineers and teachers trained abroad get a US license. ## How we built it • NVIDIA Nemotron on Nebius Token Factory does all the reasoning, with each job routed to the right size: Nemotron Nano turns the form and resume into a profile, Nemotron Super suggests goals and extracts the rules from search results, and Nemotron Ultra plans the steps. • Tavily does three jobs: it searches only government sites for rules (USCIS, DHS, DOL, the Federal Register), reads the top official pages in full with Tavily Extract, and finds this month's news. • Code handles what must never be wrong. Key dates are calculated from the F-1 regulations (8 CFR 214.2(f)). Fees and recent rule changes are checked by hand against official pages. Every step's link is verified against real search results, and costs and timelines are totaled in code. • Stack: FastAPI with streaming, React, Vite and Tailwind, SQLite for caching, Docker, deployed on Render. • Testing: unit tests, plus a script that runs six very different students through the real agent and checks each plan automatically (real fees only, correct step order, no STEM steps for a non-STEM degree, proposed rules never shown as in effect). ## Challenges we ran into Early roadmaps looked right but weren't: the model invented a $410 filing fee and put steps in the wrong order. I learned that a better prompt is not enough. The fix was to take guessing away from the model: dates calculated in code, a hand-checked list of key facts, rule searches restricted to government sites, and an automatic check of every link and every plan. Keeping up with policy was the other challenge. Several rules changed or were challenged in court during the hackathon itself, so each one needed a clear status, not just a summary. Cost mattered too: a public demo could spend real money. So the app has daily caps, a per-visitor rate limit, caching, and a switch that replays saved runs. ## Accomplishments that we're proud of • On the latest runs, every visa step links to an official source. • All six test students, from a non-STEM MBA to a PhD aiming for a university job, pass every automatic check. • It correctly handles details many guides get wrong, such as a bachelor's student not being able to enter the H-1B lottery before graduating. ## What we learned The hard part of an AI agent is not the answer, it's knowing which parts the model should never decide. Splitting the work between models of different sizes, live search and plain code made Bridgepath faster, cheaper and far more accurate than one big prompt. ## What's next for Bridgepath • Alert students when a rule that affects their saved roadmap changes. • Show real sponsorship numbers for each employer from public H-1B data. • Cover more paths, such as cap-exempt employers, O-1A and EB-2 NIW. • Partner with university international offices so DSOs can share Bridgepath with their students.

Built With

Share this project:

Updates

Submission history