LeadFinder AI

• 

  LeadFinder AI

Inspiration

As a non-technical builder at Agemo, I frequently connect with founders, investors, and potential customers. A recurring frustration I’ve heard is that researching new contacts can take 20–30 minutes per person — jumping between LinkedIn, Google, and Crunchbase to understand who someone is and how they fit.

People need to:

Identify the right customers

Find promising startups to invest in

Discover relevant people to invite to professional events

When I saw how Perplexity’s real-time search, reasoning models, and the Sonar Finance vertical worked together, it clicked — this could replace an entire stack of costly CRM tools. Traditional platforms miss or misread financial data; Perplexity nails it.

The idea: Transform passive LinkedIn connections into active business intelligence powered by Perplexity’s AI reasoning.

What It Does

LeadFinder AI turns simple inputs — a name and a company — into actionable business intelligence for founders, investors, and network builders.

Input: Name + company (or a LinkedIn connections CSV) Output (in ~25 seconds):

✅ Verified professional email (97% confidence)

✅ LinkedIn and company URLs

✅ Full company financials: revenue, valuation, funding, employee count

✅ Job title and professional bio

✅ Headquarters and company website

Two Modes:

Single Search: Prep for meetings with instant research

CSV Bulk Upload: Enrich 500+ LinkedIn contacts overnight

Use Cases:

VCs: Research founders and portfolio companies

SMBs: Identify decision-makers and warm introductions

Sales Teams: Qualify leads with verified financial data

Networkers: Unlock valuable connections from existing networks

It’s like having a 24/7 research assistant — 72× faster than doing it manually.

How I Built It

Architecture: 4-API Parallel Pipeline on the CodeWords platform All APIs run simultaneously using asyncio.gather() for maximum efficiency.

Perplexity sonar-reasoning-pro

Retrieves bios, roles, and company context

Handles ambiguous names via multi-step reasoning

Perplexity sonar-pro (Finance Vertical)

Extracts revenue, valuation, and funding data

Pulls structured financials for both public and private companies

Gemini-2.5-flash (LLM Extraction)

Replaces unreliable regex extraction with structured LLM parsing using Pydantic models

Delivers 100% accuracy extracting true metrics (e.g., employee count vs. “revenue per employee”)

SearchAPI (Google Search)

Finds LinkedIn URLs via site:linkedin.com/in queries

Solves Perplexity’s LinkedIn access limitation

Hunter.io

Finds professional emails with 96–97% confidence

Tech Stack:

Backend: Python + FastAPI (async architecture)

Orchestration: CodeWords

Data Extraction: Gemini structured outputs

Frontend: Optional v0.dev React/Next.js interface

Key Decisions:

Async parallel execution (4 concurrent APIs)

LLM-based parsing for contextual accuracy

Client-side CSV parsing with Papa Parse

Smart rate limiting for bulk requests

Challenges

LinkedIn Data Access

Issue: Perplexity can’t scrape LinkedIn

Fix: Switched to Google Search API

Result: 80%+ success rate in finding LinkedIn URLs

Financial Data Accuracy

Issue: Regex confused metrics (e.g., revenue vs. employee count)

Fix: Used Gemini’s structured parsing

Result: 100% accuracy with contextual awareness

CSV Upload (v0.dev)

Issue: No /upload endpoint on CodeWords

Fix: Implemented client-side CSV parsing

Result: Real-time enrichment progress

Conflicting Data Across APIs

Fix: Validation logic filters unrealistic data (>100K employees)

Result: Gemini resolved conflicts entirely

Performance Optimization

Fix: Parallel execution reduced search time from 100+ seconds to ~25

Accomplishments

First to Use LLM Extraction with Sonar Finance

Achieved perfect data accuracy from unstructured markdown

Could redefine financial data extraction methods

97% Email Accuracy at Scale

Tested on 50+ contacts from startups to enterprises

Multi-Model Perplexity Integration

Combined sonar-reasoning-pro + sonar-pro finance effectively

4-API Parallel Execution

Fully asynchronous, production-ready architecture

Comprehensive Data Enrichment

Returns verified emails, LinkedIn URLs, and financials

Adds contextual info (e.g., “$20B as of Sept 2025”)

Built and Validated in 48 Hours

End-to-end MVP ready with error handling, logging, and bulk upload

Honest and Transparent Data Handling

Returns null when uncertain — no fabricated data

What I Learned

From Perplexity:

Sonar Finance’s markdown-rich output is powerful but underused

Multi-model workflows outperform one-size-fits-all LLMs

API ecosystems work best when specialized models collaborate

Technically:

LLMs outperform regex for structured data extraction

Parallel orchestration cuts execution time by 4×

Validation and semantic understanding ensure credible results

From Product-Building:

Solve high-friction workflows first — that’s where value lies

Prioritize accuracy and honesty over speed

Simplicity and focus win hackathons

What’s Next for LeadFinder AI

Beta Launch: Onboard 50 early users (VCs, recruiters, and sales teams)

Internal Adoption: Use LeadFinder within Agemo for contact research

Future Enhancements:

Real-time CRM sync

Personalized insights (e.g., warm intros via mutual connections)

API access for integrations

Built With

• codewords.ai
• gemini
• hunter.io
• perplexity
• python
• v0