SpeedSearch

Indeed for university research, with contextual matching instead of keyword overlap. A student installs the SpeedSearch Chrome extension and runs a small local backend. When they visit any university research or faculty page, the extension reads it, a local LLM extracts the discrete opportunities, and each one is ranked against the student's own portfolio by the shape of the work rather than a bag of skills.

No data leaves the machine. The extension talks only to the local backend at http://localhost:5000, which talks only to Ollama at http://localhost:11434.

How it works

  Chrome tab  ->  content.js (scrape)  ->  background.js  ->  localhost:5000
                                                                   |
                                                                   v
                                                              Ollama (local LLM)
                                                                   |
                                       +---------------------------+
                                       |
                                       v
          1. Extract opportunity cards from the page
          2. Score each card vs. the student's saved portfolio
          3. Persist to SQLite
          4. Every 2 days, email each professor asking whether
             the listed opportunity is still open (one-click YES / NO)

The match prompt is designed to reward transferable depth: a student whose portfolio shows they built a browser extension from scratch will score high on any project that calls for browser-API work, even if the application domains differ. A student who only listed a skill will not.

Requirements

• Python 3.10 or newer
• Google Chrome (or any Chromium-based browser that supports MV3 unpacked extensions)
• About 4 GB of free disk for the default local model
• macOS or Linux (Windows via WSL works too)

Install and run

git clone https://github.com/ChimaOzonwoye/speedsearch.git
cd speedsearch
./setup.sh      # creates venv, installs deps, installs Ollama, pulls llama3.2:3b
./run.sh        # starts Ollama (if needed) and the Flask backend on :5000

Then load the extension:

1. Open chrome://extensions in Chrome.
2. Toggle Developer mode on (top-right).
3. Click Load unpacked and choose the extension/ folder in this repo.
4. Pin the SpeedSearch icon to the toolbar.
5. Click the icon, open Edit your profile, and either paste a description of your work or upload a portfolio PDF. The local model extracts what you have actually built; that becomes the input to every future match.

From that point on, open any research or faculty page at your university and click the SpeedSearch icon. The badge shows the top match score for the page; the popup shows each opportunity, the difficulty, the required capabilities, and a one-sentence reason for the score.

Configuration

setup.sh writes a .env file you can edit:

Variable	Purpose
OLLAMA_MODEL	Which Ollama model to use. Default llama3.2:3b.
PING_DAYS	How often to email professors. Default 2.
SMTP_*	Optional SMTP credentials. Leave blank to log-only.
APP_HOST	Base URL the YES / NO email links point at.

Larger models (llama3.1:8b, qwen2.5:7b) give noticeably better match reasoning on machines with enough RAM.

Project layout

app.py                Flask backend, Ollama calls, SQLite, scheduler
extension/
  manifest.json       MV3 manifest
  background.js       service worker, per-tab state, badge
  content.js          page scraper + research-page heuristic
  popup.html / .js    toolbar popup that shows matches
  options.html / .js  profile setup and backend selection
  icons/              toolbar icons
setup.sh              installs dependencies and Ollama
run.sh                starts Ollama and the backend
requirements.txt      Python dependencies
LICENSE               GNU GPL v3

License

SpeedSearch is released under the GNU General Public License, version 3. The text of the GPL itself may not be modified.

Every source file in this repository carries the standard GPL v3 header. Anyone is free to study, modify, and redistribute SpeedSearch under the terms of the GPL; any derivative work must be released under the same license and with source available.

If you want to use SpeedSearch inside a proprietary product, or you need a license without GPL-style obligations (for example an institutional deployment that needs vendor indemnification), contact the founding team directly. SpeedSearch is offered under dual licensing: free for the GPL-compliant open-source community, and available under separate commercial terms for institutions.

Founders: Aadharsh Sakkaravarthy, Ted Erdos, Ali Salama, Chima Ozonwoye.

Warranty disclaimer. As stated in the GPL, SpeedSearch is provided "as is" without any warranty. Any compliance obligations that apply to a deployment (FERPA, ADA, NYC bias-audit requirements, university copyright disclaimers) must be addressed in a separate agreement, not by the license.