PDF Viewer

BUILDER'S SANDBOX

Build This Paper

Use an AI coding agent to implement this research.

OpenAI Codex
OpenAI CodexAI Agent

Lightweight coding agent in your terminal.

Claude Code
Claude CodeAI Agent

Agentic coding tool for terminal workflows.

AntiGravity IDE
AntiGravity IDEScaffolding

AI agent mindset installer and workflow scaffolder.

Cursor
CursorIDE

AI-first code editor built on VS Code.

VS Code
VS CodeIDE

Free, open-source editor by Microsoft.

MVP Investment

$9K - $12K
6-10 weeks
Engineering
$8,000
Cloud Hosting
$240
SaaS Stack
$300
Domain & Legal
$100

6mo ROI

0.5-1.5x

3yr ROI

5-12x

Computer vision products require more validation time. Hardware integrations may slow early revenue, but $100K+ deals at 3yr are common.

Talent Scout

S

Shreshth Rajan

Harvard College, Cambridge, MA, USA

Find Similar Experts

Vulnerability experts on LinkedIn & GitHub

References (5)

[1]
Multi-Agent Code Verification via Information Theory
2025Shreshth Rajan
[2]
An Empirical Study of Vulnerabilities in Python Packages and Their Detection
2025Haowei Quan, Junjie Wang et al.
[3]
Self-Consistency Improves Chain of Thought Reasoning in Language Models
2022Xuezhi Wang, Jason Wei et al.
[4]
Billion-Scale Similarity Search with GPUs
2017Jeff Johnson, Matthijs Douze et al.
[5]
The Tree-Sitter
2006M. Bennington-Davis

Founder's Pitch

"Develop a multi-agent system for zero-shot vulnerability detection surpassing fine-tuned models in recall."

Vulnerability DetectionScore: 7View PDF ↗

Commercial Viability Breakdown

0-10 scale

High Potential

1/4 signals

2.5

Quick Build

4/4 signals

10

Series A Potential

2/4 signals

5

Sources used for this analysis

arXiv Paper

Full-text PDF analysis of the research paper

GitHub Repository

Code availability, stars, and contributor activity

Citation Network

Semantic Scholar citations and co-citation patterns

Community Predictions

Crowd-sourced unicorn probability assessments

Analysis model: GPT-4o · Last scored: 2/19/2026

🔭 Research Neighborhood

Generating constellation...

~3-8 seconds

Why It Matters

Vulnerability detection systems that operate without extensive fine-tuning or labeled data reduce the cost and complexity of securing software, critical in fast-paced development environments.

Product Angle

Package MULTIVER as a security-as-a-service offering that integrates into existing CI/CD pipelines, providing real-time vulnerability analysis and reporting.

Disruption

Replaces the need for extensive labeled training data and complex fine-tuning in existing vulnerability detection methods, making security analysis more accessible and affordable.

Product Opportunity

The cybersecurity market, particularly within software development, where a robust security layer can prevent costly breaches and application downtime. Businesses pay for effective security solutions that lower the risk of software vulnerabilities.

Use Case Idea

Use MULTIVER as a security audit tool within software development pipelines, especially for companies that cannot afford extensive training data or systems.

Science

The paper introduces MULTIVER, a multi-agent system that uses a zero-shot approach involving multiple specialized agents (security, correctness, performance, style) working in parallel and combining their outputs through ensemble voting. This allows detection of software vulnerabilities across multiple dimensions without the need for labeled training data, achieving high recall by leveraging union voting that maximizes detection at the cost of increased false positives.

Method & Eval

The method was tested on benchmarks like PyVul and SecurityEval, achieving 82.7% recall on PyVul, surpassing fine-tuned models like GPT-3.5 in recall. Ablation studies showed that each agent contributed significantly to the recall, and retrieval augmentation added precision.

Caveats

High false positive rate (85% FPR) which could lead to excess manual reviews and decreased efficiency in a production setting. The system is costly per sample and not suitable for real-time CI/CD gating.

Author Intelligence

Shreshth Rajan

LEAD
Harvard College, Cambridge, MA, USA
shreshthrajan@college.harvard.edu