Skip to content
DEATHGATE01Public

About

Explainable lead-scoring pipeline for biotech outreach using LinkedIn, PubMed & funding signals.

Topics

Resources

Stars

0 stars

Watchers

0 watching

Forks

Latest commit

 

History

2 Commits

Folders and files

Repository files navigation

BioLeadIQ

Reproducible Lead Intelligence for 3D In-Vitro Therapeutics


Overview

BioLeadIQ is a lightweight lead-intelligence system designed to help business development teams prioritize who to speak to first when selling or collaborating around 3D in-vitro models for therapeutic research.

Instead of collecting large volumes of unverified data, the system focuses on explainable decision signals and produces a ranked, reproducible output that can be inspected directly in a spreadsheet.

The emphasis is on clarity, reproducibility, and business relevance, not automation for its own sake.


What Problem Does This Solve?

Business development teams often have access to many potential contacts but limited insight into:

  • Who is scientifically active in the relevant area
  • Who is likely to have budget
  • Who fits the right seniority and role

BioLeadIQ consolidates identity, scientific intent, and budget readiness into a single, ranked view, allowing teams to prioritize outreach instead of manually triaging long contact lists.


Design Principles

This project was built with the following principles:

  • Reproducibility over live scraping
  • Explainability over black-box scoring
  • Business signals over raw data volume

Every design decision in the system reflects these priorities.


Data Sources and Why They Were Chosen

LinkedIn (Identification)

For the demo, LinkedIn profile data is ingested from a static CSV (data/linkedin_profiles.csv) containing:

  • Name
  • Title
  • Company
  • Location
  • LinkedIn URL

This approach ensures the demo is easy to verify and reproduce. In a production setup, the same schema could be populated via providers such as Proxycurl or Apify without changing the rest of the pipeline.


PubMed (Scientific Intent)

Scientific intent is inferred using publication signals from PubMed:

  • Recency of publications (last 24 months)
  • Presence of domain-specific keywords (e.g., toxicology, hepatic models, DILI)

For reproducibility, the demo uses a deterministic, cached mapping inside enrichment/pubmed_client.py. An optional live PubMed lookup via NCBI E-utilities is supported but disabled by default to avoid network and rate-limit variability.

This allows reviewers to see consistent results while still demonstrating how live enrichment would work.


Funding Data (Budget Signal)

Company funding data is sourced from a static JSON file (data/funding_data.json) containing:

  • Funding stage
  • Last funding year

This captures a coarse but practical signal of budget readiness. The design intentionally keeps this layer simple; in production, it can be replaced with a Crunchbase or PitchBook integration.


How the Scoring Works

Each lead is assigned a Propensity Score (0–100) using a transparent, rule-based model:

score = role_fit * 30
      + scientific_intent * 40
      + company_funding * 20
      + location_hub * 10

The weights reflect typical business priorities:

  • Scientific activity is the strongest indicator of near-term relevance
  • Role seniority and fit determine decision-making power
  • Funding stage indicates ability to purchase or adopt new tools
  • Location acts as a secondary signal for ecosystem maturity

All scoring logic is implemented in scoring/scoring_rules.py and is fully inspectable.


Output and Demo Artifact

Running the pipeline via:

python main.py

always produces a ranked output.csv.

If Google Sheets credentials are provided, the same results are also written to a Google Sheet, where they can be:

  • Sorted
  • Filtered
  • Reviewed without running code

The Google Sheet is treated as the primary demo artifact, while the code remains the source of truth.


Reproducibility and Tradeoffs

This demo deliberately avoids:

  • Live LinkedIn scraping
  • Black-box ML or LLM scoring
  • Non-deterministic pipelines

These choices ensure:

  • Results can be verified quickly
  • Reviewers can inspect logic without setup friction
  • The system remains easy to reason about

The tradeoff is reduced data breadth, which is acceptable for a demo focused on decision logic rather than scale.


Limitations and Extensions

  • LinkedIn and funding data are static samples; coverage depends on input quality
  • PubMed live lookup is optional and rate-limited
  • Funding stages are coarse and may miss nuance
  • Location hub detection is heuristic and region-specific

These limitations are intentional for the demo and can be addressed in a production deployment.


Summary

BioLeadIQ demonstrates how a small, well-designed system can turn fragmented public signals into a clear, ranked decision tool.

The goal is not to replace CRM systems or crawlers, but to show how transparent logic and scientific context can meaningfully improve lead prioritization.

About

Explainable lead-scoring pipeline for biotech outreach using LinkedIn, PubMed & funding signals.

Topics

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages