I hate how long it takes to prepare for conferences. Hours and hours of identifying the right companies among thousands of attendees, cross-referencing them with my thesis, and trying to figure out which 30 are actually worth my time. So I built this.
VC Conference Scout takes a conference's attendee list (a URL to scrape, or a CSV you upload), runs every company through Claude using your investment thesis, and gives you back a filterable, scored shortlist with funding stage, location, founding year, and a one-sentence "why it's relevant" — for hundreds of companies in a few minutes.
It's a self-hosted tool. You bring your own Anthropic API key. Your data never leaves your machine.
- Scrape an attending-companies page (or load companies from a CSV/TXT file).
- Pre-filter out the obviously-not-a-startup names (Fortune 500, big banks, consulting firms, VCs, associations).
- Classify & score every remaining company against your thesis using Claude. Each company gets a 0–10 relevance score and a one-line reason.
- Enrich the top 30 candidates via Claude's web search tool: founded year, funding raised, stage, HQ, what they actually do, why they're innovative.
- Auto-categorize the final shortlist into 8–12 categories tailored to this specific dataset (or define your own).
- Browse in a clean web UI: filter by category, stage, funding, score range; star favorites; export starred companies to CSV.
The night before HumanX 2026 I had a list of 1,000+ attending companies and roughly zero hours to read about all of them. I needed:
- A way to throw out the 80% that obviously don't fit my thesis (big tech, banks, consulting firms, other VCs).
- A way to score the remaining 20% by how well they actually match what I invest in — not generic relevance, my thesis.
- A way to look up funding stage, founding year, and HQ without opening hundreds of Crunchbase tabs.
- A way to take notes and star the ones I actually wanted to meet.
A few minutes and a few dollars of Claude API spend later I had a filterable, scored shortlist with every match enriched with real descriptions. This is that tool, generalized so any VC can point it at any conference.
Requirements: Python 3.8+ and an Anthropic API key.
git clone https://github.com/arbiterFF/vc-conference-scout.git
cd vc-conference-scout
# Create a virtual env (avoids the "externally-managed-environment" error
# you'll hit on modern macOS / Debian / Ubuntu if you skip this).
python3 -m venv .venv
source .venv/bin/activate # Windows: .venv\Scripts\activate
pip install -r requirements.txt
python3 server.pyThe server runs on http://localhost:8081. Open it in your browser.
- Click ⚙ Settings → paste your Anthropic API key → click Save Key.
- Write your investment thesis in the Profile field. Or paste your firm's website URL and click Extract from URL — Claude will read the site and draft a profile for you.
- Click + New Scan → give it a name → choose source:
- From URL: paste the conference's "attending companies" page URL. Note: many conference sites (Map-Your-Show, 6Connex, Webflow) load their attendee list with JavaScript and won't work with URL scraping — if your scan fails with a "couldn't find any company names" error, fall back to CSV upload.
- CSV upload: drop a CSV/TXT file. The parser auto-detects which column holds the company names (looks for a header like
name/company/exhibitoror scores each column for name-like content).
- Choose Auto-generate categories (recommended) or Manual.
- Hit Start Scan. Watch the progress bar. Browse the results when it's done.
error: externally-managed-environmentwhen runningpip install→ you skipped the venv step. Go back and runpython3 -m venv .venv && source .venv/bin/activatefirst.ModuleNotFoundError: No module named 'requests'when runningpython3 server.py→ yourpipandpython3resolved to different Python installations. Usepython3 -m pip install -r requirements.txtto be explicit.OSError: [Errno 48] Address already in use→ something else is on port 8081. Either stop that process (lsof -ti :8081 | xargs kill) or change the port inserver.py(last few lines).python3: command not found→ install Python 3 from python.org or via Homebrew (brew install python).
Stored in your browser's localStorage (never on disk on the server):
| Setting | Purpose |
|---|---|
| Anthropic API Key | Your sk-ant-... key. Used for all Claude calls. |
| VC Profile / Thesis | Free-text description of what you invest in. Injected into every classification prompt. |
Stored on disk per scan in scans/<scan-id>/config.json:
| Field | Purpose |
|---|---|
name, url, selector |
What to scrape |
profile, categories |
Scoring + bucketing rules |
auto_categorize |
If true, regenerates categories from the actual results |
source |
url or csv |
Your API key is only kept on disk for the duration of a scan, then automatically stripped from the config file when the scan completes.
vc-conference-scout/
├── scout.py # Pipeline: scrape → filter → classify → enrich → categorize
├── server.py # Multi-scan REST API + static SPA host
├── index.html # Single-page app: home, new scan, settings, scan detail
└── scans/
└── <scan-id>/
├── config.json
├── companies.json (only when CSV-uploaded)
├── results.json
├── starred.json
├── progress.json
└── scout.log
Each scan runs as a detached subprocess writing progress.json for the UI to poll. Multiple scans can run in parallel.
If you give it a URL, it tries two strategies:
- CSS selector — if you provide one (under "Advanced"), it's used directly.
- Auto-detect — finds the parent element on the page whose direct children look most like a list (≥30 children, mostly short text, no
<br>-separated noise). This works on most conference attending-companies pages without any tweaking.
If auto-detect fails, open the page in your browser, inspect the company names, and provide a CSS selector like .attendee-list .company-name.
CSV uploads skip the scraping step entirely.
That's the pipeline doing its job. It's a funnel:
N scraped
↓ pre-filter (drops obvious mega-corps, VCs, consulting firms)
↓ coarse triage (Claude reads each name, drops obvious mismatches — permissive)
↓ enrich (web search) (every survivor with relevance ≥ 5 gets a full description)
↓ score with full context (the REAL score, using the description not the name)
shortlist in the UI
Most companies at any conference are mature businesses, service providers, banks, VCs, or simply unrelated to your thesis. The pipeline drops them along the way. The UI's score slider also defaults to 5–10 — anything lower is hidden by default.
If you want a longer list, edit scout.py → batch_classify() and lower the relevance >= 5 threshold (or remove the startup|growth type filter to keep mature companies too).
Each scan with ~3,000 attendees costs roughly $5–10 in Claude API spend:
- Classification: ~17 batches × ~$0.10 each
- Enrichment of top 30 (with web search): ~$0.10–0.20 each
- Auto-categorize (if enabled): ~$0.50
You can lower this by enriching fewer companies or skipping the auto-categorize step.
- Hosted SaaS version with sign-in (no clone-and-run required)
- Background re-scoring with enriched context (
rescore.pyfrom the original Manifest version) - Enrich-all (not just top 30) for power users
- LinkedIn / Crunchbase enrichment
- Multi-conference dedup (track companies you've already evaluated)
- Notes per company
PRs welcome. The whole thing is ~1,500 LOC across 3 files; it's meant to be hackable.
MIT. Built because hand-scrolling thousand-company attendee lists at 1am the night before a conference is no way to live.