Founder Research Plan — People Intelligence

People Intelligence Project — Mapping the LA AI ecosystem and personal networks

1. Goal & Strategy

Objective: Map the network of people around Tim & Tara Chklovski and TenOneTen Ventures (Gil Elbaz, David Waxman, Eric, Minnie), cross-referenced with the LA AI startup ecosystem. Surface who knows whom, how they're connected, and where the warm intros live.

The approach is data-first: get real LinkedIn profile data, analyze it, and build the network from what we actually find — rather than over-engineering a product before seeing what the data supports.

2. Hard Constraint: LinkedIn Connections

LinkedIn connections are unavailable from ANY scraper

LinkedIn's Connections API requires OAuth consent from each individual user. No third-party — not Bright Data, not Apify, not anyone — can extract who someone is connected to on LinkedIn. Proxycurl (formerly the main alternative) was shut down by LinkedIn lawsuit in July 2025.

What IS available vs. what ISN'T

The Workaround: Two Complementary Methods

Data	Available?	Source	Notes
Full employment history with dates	Yes	Profile scrape	Company, title, start/end dates for every job
Education with dates	Yes	Profile scrape	School, degree, field, year ranges
Current company & title	Yes	Profile scrape	"Present" end date = currently there
Skills, about section, posts	Yes	Profile scrape	Full text content
Connection list (who they know)	No	—	Requires individual OAuth. No scraper has this.
Connection date (when connected)	Self-export only	User's own CSV export	Settings → "Download your data" → Connections.csv
Profile changes over time	Re-crawl only	Periodic re-scrape	Diff between scrapes shows job changes, new roles

Method A: Self-Export (Gold Path)

Each person exports their own LinkedIn connections:

LinkedIn → Settings → Get a copy of your data → Connections

Gives: name, email, company, position, connected date (the recency signal).

Takes 30 seconds per person. This is the only way to get actual connection lists with dates.

Method B: Inferred Network (Automated)

Scrape profiles of known contacts, then infer connections from:

Shared employer overlap — weighted by org size & time
Shared education overlap — weighted by year overlap
Co-investment — from Crunchbase/SEC (free)
Board co-membership — from SEC DEF 14A (free)

Edge weighting: 50-person startup overlap = strong. Google overlap = weak (unless same team/location).

3. Phase 1: Personal Network Extraction

Data Captured Per Contact

Priority	Person	Affiliation	Method	Expected Contacts
1	Tim Chklovski	Personal	Self-export Connections.csv	500-2,000+
1	Tara Chklovski	Personal	Self-export Connections.csv	500-2,000+
2	Gil Elbaz	TenOneTen Ventures	Ask for export (or infer from profile scrape)	1,000-5,000+
3	David Waxman	TenOneTen Ventures	Ask for export (or infer)	1,000-3,000+
4	Eric	TenOneTen Ventures	Ask for export (or infer)	500-2,000
5	Minnie	TenOneTen Ventures	Ask for export (or infer)	500-2,000

Every contact record in SQLite includes:

Field	Source	Example
`name`	CSV / scrape	Jane Smith
`company`	CSV / scrape	OpenAI
`title`	CSV / scrape	VP Engineering
`email`	CSV export only	jane@openai.com
`connected_date`	CSV export only	2024-03-15
`whose_contact`	Import metadata	gil_elbaz
`extraction_method`	Import metadata	linkedin_csv_export
`extraction_date`	Import metadata	2026-02-06
`linkedin_url`	CSV / scrape	linkedin.com/in/janesmith

Processing Pipeline

Connections CSVs Profile Scrapes | | v v Import to SQLite Import to SQLite (name, company, title, (full employment history, email, connected_date, education, skills, posts) whose_contact, method) | | +----------+---------------------+ | v Deduplicate across contact lists (who do Tim AND Gil both know?) | v Scrape unique profiles (~200-500 profiles via BD/Apify) | v Build employment/education graph (shared history = inferred connections)

4. Phase 2: LA AI Ecosystem

Cross-reference personal contacts with the 1,343 AI companies identified in Greater Los Angeles.

Dataset Summary

Top LA AI Companies by Funding

Notable LA AI Companies (Web Research)

By Sub-Region

Size Distribution

5. Phase 3: Network Graph & Analysis

#	Company	Focus	Funding	Employees	Founded
1	Faraday Future	AI electric vehicles	$4.08B	1,001-5,000	2014
2	Alteryx	Data science & analytics	$1.41B	1,001-5,000	2011
3	Relativity Space	3D-printed rockets + AI	$1.36B	501-1,000	2015
4	System1	Predictive analytics	$920M	101-250	2013
5	Vercel	Frontend cloud + AI	$563M	251-500	2015
6	FLYR	AI for airlines	$482M	501-1,000	2013
7	Veritone	Enterprise AI platform	$393M	501-1,000	2014
8	Cylance	AI cybersecurity	$297M	501-1,000	2012
9	Genies	AI avatar tech	$267M	101-250	2011
10	Reveleer	Healthcare NLP	$209M	101-250	2009

Company	Location	Focus	Notes
Anduril Industries	Costa Mesa	Defense AI, autonomous systems	$6.84B raised, $78B valuation. Palmer Luckey. ~6K employees.
HeyGen	Los Angeles	AI video generation	$60M Series A, ~$95M ARR. Hot growth.
GrayMatter Robotics	Carson	AI manufacturing robots	$45M Series B. BuiltInLA "Best Startup" 2026.
Avenda Health	Culver City	Cancer detection AI (FDA-cleared)	Multi-modal AI for 3D cancer maps.
Beyond Limits	Glendale	Industrial AI	NASA/JPL heritage. 45 space technologies adapted.
Virtualitics	Pasadena	AI data analytics	Caltech spinout. DOD + Fortune 500.
Snap Inc.	Santa Monica	AR/AI, computer vision	Public. 5K employees. Heavy AI investment.
Metropolis	Los Angeles	Computer vision payments	Largest parking network in N. America. 50M+ customers.

Area	Notable Companies
Santa Monica	Snap, Axle Health, GumGum, Zeitview, Emerge Tools
Venice / Mar Vista	AE Studio
Culver City	Avenda Health
Playa Vista	Google campus, various startups
El Segundo	HABIT (YC), Dogtown Media, Plug&Play AI
Pasadena	Virtualitics, Deep 6 AI, Supplyframe, Caltech/JPL ecosystem
Glendale	Beyond Limits (NASA/JPL heritage)
Costa Mesa	Anduril Industries

Employee Range	Count	% of Total
1-10	596	44%
11-50	477	36%
51-100	83	6%
101-250	44	3%
251-500	18	1%
501+	18	1%

Who do Tim and Gil both know? (intersection of contact lists)
Which contacts work at LA AI companies? (cross-reference with 1,343 companies)
Who's connected to whom via shared history? (employment/education overlap graph)
Who are the "super-connectors"? (people appearing in multiple contact lists)
Warm intro paths: Tim → [shared contact] → [target founder/VC]

Edge Types & Strength

Connection Type	Strength	Source	Recency Signal?
Direct contact (from CSV export)	Known	LinkedIn export	Yes — `connected_date` in CSV
Currently at same company	Strong	Profile scrape ("Present")	Yes — ongoing
Shared employer, recent (<2yr ago)	Strong	Profile scrape (dates)	Yes — date ranges
Shared employer, older (2-5yr)	Medium	Profile scrape (dates)	Yes — computable
Shared employer, historical (>5yr)	Weak	Profile scrape (dates)	Yes — computable
Shared education (overlapping years)	Medium	Profile scrape (dates)	Historical only
Co-investment in same company	Strong	Crunchbase / SEC	Yes — funding dates
Board co-membership	Strong	SEC DEF 14A	Yes — filing dates

Critical: Edge Weighting by Org Size

Two people who both worked at a 50-person startup = very likely know each other.

Two people who both worked at Google (200K employees) = probably never met unless we can confirm same team/office/dates.

The graph must weight edges by organization size at the time of overlap to avoid false positives.

6. LinkedIn Scraping: Options & Alternatives

Provider Comparison

Recommendation

Provider	Cost/Profile	Reliability	Setup	Best For	Notes
Bright Data	$0.0015-0.0025	Best	API key + MCP	1,000+ profiles at scale	Largest proxy network. 544M+ profiles. MCP already installed in rivus. Promo: `APIS25` (25% off 6mo).
Apify	$0.01-0.03	Good	Apify account + actor	Quick batches of 100-1,000	Easiest to start. "LinkedIn Profile Scraper" actor. Pay per compute unit. No proxy management needed.
ScrapingBee	$0.005-0.01	Decent	API key	Medium batches	JS rendering support. LinkedIn-specific endpoint.
People Data Labs	$0.004-0.28	Aggregated	API key	Enrichment, not real-time	Pre-compiled data. Free tier: 100 records/mo. May be months stale.
PhantomBuster	~$0.03/lead	Risky	Your LinkedIn login	Avoid	Uses YOUR LinkedIn credentials. Account ban risk. Not recommended.
DIY (Playwright + proxies)	~$0.001	Fragile	Code + proxy subscription	Cheapest but most work	Residential proxies ~$0.01/hr. LinkedIn actively blocks. High maintenance.

For our initial batch (~200-500 profiles):

Bright Data is ideal — MCP is already installed, cheapest at scale, most reliable
Apify as backup — easier to get started, good for quick tests, ~10x more expensive per profile
Test 10 profiles first with either provider before running bulk
Set hard spend limits ($5 cap for Phase 1)

Cost at 500 profiles: Bright Data = $0.75-1.25 | Apify = $5-15

What None of Them Can Do

No LinkedIn scraper — paid or free — can extract:

A person's connections list (who they're connected to)
When two people became connected
Mutual connection counts between two people
Connection activity (who commented on whose posts)

These all require the person's own OAuth token. The only way to get connection data is the self-export CSV method described above.

7. Data Schema

SQLite Tables (people.db)

contacts profiles id INTEGER PRIMARY KEY id INTEGER PRIMARY KEY name TEXT NOT NULL linkedin_url TEXT UNIQUE email TEXT name TEXT company TEXT headline TEXT title TEXT location TEXT linkedin_url TEXT about TEXT connected_date TEXT current_company TEXT whose_contact TEXT NOT NULL current_title TEXT extraction_method TEXT skills TEXT (JSON array) extraction_date TEXT posts TEXT (JSON array) created_at TIMESTAMP scraped_at TIMESTAMP scrape_source TEXT employment_history education_history id INTEGER PRIMARY KEY id INTEGER PRIMARY KEY profile_id INTEGER FK profile_id INTEGER FK company TEXT school TEXT title TEXT degree TEXT start_date TEXT field TEXT end_date TEXT start_year INTEGER location TEXT end_year INTEGER is_current BOOLEAN edges (inferred connections) la_ai_companies id INTEGER PRIMARY KEY id INTEGER PRIMARY KEY person_a_id INTEGER FK name TEXT person_b_id INTEGER FK description TEXT edge_type TEXT categories TEXT strength REAL (0-1) location TEXT shared_entity TEXT funding_total REAL overlap_start TEXT employees TEXT overlap_end TEXT website TEXT org_size_at_time INTEGER linkedin TEXT notes TEXT founded TEXT

8. Parallel Execution Timeline

Day 0 — All parallel

Ask Tim & Tara to export LinkedIn Connections.csv (30 seconds each)
Ask Gil, David, Eric, Minnie (TenOneTen) for same export
Build SQLite schema & import pipeline for connections CSVs
Import LA AI companies (1,343) into la_ai_companies table

Day 1 — CSVs arrive + scrape prep

Import CSVs as they arrive (each with whose_contact, extraction_method, extraction_date)
Deduplicate contacts across lists — identify who appears in multiple lists
Cross-reference with LA AI companies — which contacts work at the 1,343 companies?
Queue LinkedIn scrape of unique contacts (~200-500 profiles)

Day 2 — Scrape + first results

Run LinkedIn profile scrape via Bright Data (test 10, then bulk)
Extract employment + education history from scraped profiles
Begin graph build — shared employer/education edges from first batch

Day 3-4 — Graph & analysis

Complete graph build — all edges with strength weighting
Run queries: Who do Tim & Gil both know? Super-connectors? Warm intro paths?
Generate network visualizations (pyvis / Plotly)
Cross-reference with LA AI ecosystem data

Day 5 — UI & demo

Gradio UI showing network map, contact cards, reachability
Interactive queries: "Who can intro me to [person at AI company X]?"

9. Budget

10. Risks & Mitigations

11. Founder Evaluator (Future Phase)

Item	Cost	Details
LinkedIn Connections CSV exports	$0	Self-export by each person
LA AI companies data (Crunchbase)	$0	Already downloaded (1,343 companies)
Free data sources (Forbes, SEC, Signal, OpenVC)	$0	Already collected (60K+ records)
Profile scrape: 200-500 contacts (Bright Data)	$0.50-1.25	At $0.0025/profile
Profile scrape: 200-500 contacts (Apify alt)	$2-15	At $0.01-0.03/profile
Crunchbase/SEC cross-reference	$0	Existing data
Total (Bright Data path)	$0.50-1.25
Total (Apify path)	$2-15

Risk	Impact	Mitigation
People don't export their CSVs promptly	Medium	Start with whoever exports first. Build pipeline with Tim/Tara's data, add others as they arrive.
Noisy inferred network (everyone went to Stanford)	Medium	Weight edges by org size at time of overlap. Small company = strong. Large = weak unless same team.
Bright Data cost overrun	Low	Hard API spend limit at $5. Test 10 profiles before bulk. Initial batch is tiny (~$1).
Identity resolution (duplicate Smiths)	Medium	Match on (Name + Company) or (Name + School). If uncertain, keep both — don't merge aggressively.
LinkedIn scraping blocked/degraded	Low	Bright Data handles anti-bot. Apify as fallback. Test first.
Data staleness over time	Low	Monthly refresh of key profiles ($0.0005/profile). Track `scraped_at` date per record.

Profile Card

Background, trajectory, education, prior exits, technical depth. Auto-generated from scraped profile data.

Network Map

Interactive visualization of connections to VCs, founders, and LA AI companies. Pyvis / Plotly.

Founder Score (MVP: 3 dims)

Prior startup success, network quality, technical depth. Backtested against YC acceptances.

Warm Intro Paths

"Tim → [shared contact] → [target person]" with relationship strength and recency.

Full spec: projects/people/README.md | Data: projects/people/data/ | Source: rivus

Founder & Network Research Plan