Match VC

Inspiration

The seed of Match VC came from living the pain of fundraising on both sides of the table.

Founders on our team spent weeks wrestling with giant spreadsheets, outdated data brokers, and cold-email spray campaigns that rarely converted.
Angel & VC friends told us their inboxes were flooded with off-thesis pitches and “just checking in” nudges that took hours to triage.

We asked ourselves “What if the first thing a startup saw was a list of the right investors and the first thing a VC saw was a shortlist of on-thesis companies?” That conviction – plus frustration at pay-walled databases and boiler-plate CRMs – birthed Match VC.

What it does

Match VC is a two-sided, AI-powered fundraising workspace:

For Startups	For VCs
🔍 AI matching - instantly ranks investors by fit, cheque size & recent activity	📥 Curated deal flow - only startups that meet declared thesis & ticket stage
✉️ 1-click pitch generation - personalized subject, email, deck points, follow-ups	🗂 Batch triage - swipe or hot-key through deals, auto-sync notes to CRM
📊 Responsiveness metrics - see which VCs reply fastest	📈 Portfolio & market analytics - IRR snapshots, sector trends, gap analysis
🕸 Warm-intro graph - suggests strongest path via Gmail/LinkedIn contacts	💬 Reputation score - build public standing by fast, constructive responses
🏆 Gamified achievements - points, badges, leaderboards to drive engagement	🔔 Smart alerts - new company matches, momentum spikes, doc-room uploads

Everything is wrapped in a clean Next.js 13 UI, responsive from desktop to mobile, with dark/light modes and silky Framer Motion animations.

How we built it

Architecture – Next.js 13.5 + React/TypeScript front end, serverless API layer (Cloudflare Workers), PostgreSQL for transactional data, Redis for 5-min cache TTL, Pinecone for vector search embeddings.
Data ingestion & cleansing – real-time pull from a custom VC data API (150 k+ investor records), plus Hunter.io & Debounce for e-mail verification. FastText used for country/sector normalization.
AI services
- Matching → OpenAI text-embedding-3-large builds 768-dim vectors of startup blurbs and investor theses; cosine similarity feeds the Fit Score.
- Content generation → GPT-4o API with structured prompt templates; A/B subject line tests stored for reinforcement learning later.
- Startup scoring → five-pillar weighted model (team, market, product, traction, finance) with confidence intervals.
Warm-intro graph – Google People API & LinkedIn OAuth import → Neo4j graph DB → Dijkstra to rank paths (edge weights = interaction frequency + seniority).
Gamification & analytics – Achievement engine in a standalone service worker; Mixpanel captures engagement funnels and triggers in-app nudges.
CI/CD – Turborepo monorepo, GitHub Actions, pnpm, Playwright + Storybook for visual regression, and Terraform for infra as code.

Challenges we ran into

Real-time data freshness – investor moves and fund closures happen daily. We built a delta-scraper and reconciliation queue to keep latency under 24 hours.
Email-provider trust limits – bulk sending personalized mails tripped Gmail safe-guards; we introduced staggered send-queues and deliverability scoring.
Graph privacy – hashing e-mails client-side before graph upload so raw addresses never leave the user browser; required WebCrypto and a custom protocol buffer.
Bias in AI scoring – early models over-favoured US/fintech. We re-weighted feature importance and injected synthetic under-represented data during training.
Keeping first-run UX < 3 min – many features, but the “aha!” moment had to be instant; we fought scope creep and built an empty-state wizard that seeds 40 investors automatically.

Accomplishments that we’re proud of

Sub-2-second search across 150 k investors with quality scoring and live filters.
< 3-second full pitch generation tested on 200+ prompts during internal dog-food week.
91 % email deliverability in closed beta thanks to verification & staggered queues.
80 % of beta founders booked > 3 meetings in week 1 (vs Ø 1.2 on their prior tools).
Security first – completed SOC-2 Type I readiness checklist in four weeks.

What we learned

Onboarding is product-market fit in miniature. The warm-intro graph and pre-seeded pipeline tripled user activation compared with a blank search bar.
VCs are users too. Investor retention correlated strongly with features that saved them time (batch triage, thesis toggle), not just better inbound quality.
Gamification works, but only when it feels earned. Public “Quick Responder” badges drove a +14% reply-rate; vanity badges with no behavioural tie-in were ignored.
AI ≠ magic without pristine data. Cleaning, deduping and feature engineering consumed more hours than prompt crafting – and was worth every minute.