1 of 30

PARSONS E-LAB

(01/10)

Hidden Alpha: Reframing Bias in Early-Stage Investment Decisions

Reframing AI for Equity

AN ANTI-BIAS TRAINING TOOL FOR CAPITAL ALLOCATORS

anti-bias Training

A Systems Thinking Approach to Inclusive Startup Ecosystem Design

Hidden Alpha: Reframing Bias in Early-Stage Investment Decisions

Reframing Capital: AI, Bias, and Decision Infrastructure in Early-Stage Investing

2 of 30

This project is rooted in over two years of extensive research accomplished by Professor Rhea Alexander and the Parsons Entrepreneurial Lab (E-lab).

Prof. Rhea Alexander

Head of E-Lab

Yash Sonwaney

Administrative Lead

Rudy Ofori

Technical Lead

Meet

The Team

(02/10)

PARSONS E-LAB

3 of 30

2 .5x

1%

98%

78

Opportunity Loss

(05/10)

The Problem

PARSONS E-LAB

When Melinda Gates launched Pivotal Ventures, focusing on women-led businesses, the portfolio significantly outperformed traditional VC funds in similar sectors. It turns out, female founders generate 78 cents per dollar invested vs. 31 cents for male founders—that's 2.5x better capital efficiency. They also burn 15% less capital on average, which means better economics and dealflow. However, female-founded startups currently receive under 2% of all investment.

This creates an untapped market segment of 98% that investors can capitalize on to improve their returns if they can refine their analysis and overcome their biases.

4 of 30

We aim to create an AI enabled open-source web application that enables capital allocators (generally early-career VCs, angel investors, and post-secondary MBA program students) improve portfolio performance by increasing bias awareness when evaluating startup pitches, at the earliest funding stages.

Our Solution

(06/10)

PARSONS E-LAB

WHAT

Hidden Alpha: Reframing Bias in Early-Stage Investment Decisions

See signal. Reduce noise. Capture missed returns.

5 of 30

The open-source, AI-enabled platform would be accessible online. Users login and select from a library of pre-recorded, five-minute seed-stage pitch videos from early-stage founders. Some recordings feature blind (mediated) pitches, where founders present through masked avatars with voice modulation, while others are presented natively.

After viewing a pitch, investors complete a short questionnaire and then enter an open AI-driven evaluation space, where they can “dialogue” with the AI. The system analyzes their questions and responses for potential bias or decision-making blind spots, generates a scorecard, and offers feedback. Investors can then repeat the process with additional pitches from the library.

The overarching goal is to help investors recognize and correct their own biases in order to improve capital allocation decisions, diversify their portfolios, and ultimately achieve stronger returns. This is not about hiding information—it is about reordering how information is presented so that:

promising opportunities are not overlooked

their fund doesn’t miss out

increase returns to reach alpha sooner!

(06/10)

PARSONS E-LAB

HOW

Reframing Capital: A Transparent Capital Initiative

6 of 30

Their blind auditions increased the proportion of women in orchestras by 25-46%, with the greatest impact occurring in the final rounds of auditions. For instance, the New York Philharmonic went from less than 5% women in 1970 to 45% by 2020

They implemented "Blind CV" screening for graduate recruitment and used AI to redact identifying information, resulting in 40% increase in ethnic minority candidates progressing to assessment centers

They removed names, photos, and university information from initial job applications in 2017, increasing diversity in shortlisted candidates by 30%. More candidates from non-traditional backgrounds advanced to interviews.

US Orchestras

Success

Stories (Precedents)

HSBC

BBC

(08/10)

PARSONS E-LAB

7 of 30

Goals: Better Questions. Better Bets.

  • Improving Venture Decision Quality Under Uncertainty

Addressing : Gender bias as a pricing error

  • AI, Bias, and Decision Infrastructure in Early-Stage Investing

8 of 30

Early testing ways to achieve goals

APPROACH

PHASE 1

Fall 2025

Building platform on a server

then measure impact

PHASE 2

Spring 2026

Definition & Experimentation

Platform development & mass testing

(WHERE WE ARE CURRENTLY)

9 of 30

PHASE 2

Building platform on a server

then measure impact

PARSONS E-LAB

10 of 30

Core Platform Roadmap

(08/10)

Engagement & Interaction Design

  • Design and develop user dashboard
  • Create pitch library browsing interface with filters
  • Build video player with custom controls
  • Develop assessment questionnaire interface
  • Create AI chat interface for Q&A simulation
  • Implement progress tracking and history views

Assessment System

  • Design standardized questionnaire framework
  • Build question bank database
  • Develop response collection and storage system
  • Create scoring and evaluation logic
  • Implement randomization for question order (if applicable)
  • Build results display interface

AI Integration - Core Functionality

  • Select and configure AI model/API
  • Build prompt engineering framework for:

Founder persona simulation

Bias detection algorithms

Response analysis

  • Develop content management system (conversation history)
  • Create fallback mechanisms for AI failures
  • Implement rate limiting and cost controls

1

2

3

11 of 30

AI Model responds with improvements and suggestions on reframing language to be less biased

FRONT-END

BACK-END

How It Works

Founder pitches are recorded and stored in the platform cloud network

Brofolio analysis responses, generates scores and suggestions on reframing and change of mindset through API

Investor watches anonymized and non-anonymized startup pitches

Naturally react, ask clarifying questions and judge founders

Answer perception survey about investment size, founder perception

1

2

3

4

5

6

12 of 30

Video Walkthrough

13 of 30

How it Works

  1. Investors select from a library of video pitches without decks from founders at the pre-seed/seed stage, some of which are anonymized.
  2. Watches two video pitches by founders
    1. Condition A (Anonymized): No founder photos, 5min talking head video using an avatar and voice modulator, Gender-neutral language, sanitized bios
    2. Condition B (Traditional): 5min “talking head” video, live-action on-camera founder pitching at seed/pre-seed stage
  3. Investor shares their responses to a questionnaire and submits reactions and questions via text input into the AI
  4. Behind-the-scenes AI analysis (using trained AI model API)
    • Each question is analyzed for: Promotional focus (growth, opportunities, strengths, vision) and Preventative focus (risks, concerns, obstacles)
    • Bias indicators (disproportionate scrutiny, personal attribute vs. business questions)
  5. Investor completes a comprehensive survey measuring:
    • Investment decisions: Likelihood (1-7), Yes/No/Unsure, check size, portfolio allocation
    • Founder perceptions: Confidence, knowledge, resilience, leadership, data presentation, coachability, Q&A effectiveness
    • Risk assessments: Market, execution, team, technology, funding risks
    • Qualitative responses: Strongest/weakest aspects, additional info needed, founder-opportunity fit, desired co-founder attributes, 3-word founder description
  6. BroFolio Bias Report is presented with:
    • Overall and per-question bias score
    • Semi-mean humor opening (attention-grabbing roast)
    • Warm guidance (constructive coaching)
    • Visual charts showing promotional vs. preventative ratio
    • Question breakdown with classifications
    • Actionable recommendations

Designed with behavioral economists and neuroscientists to create self-awareness and an open mindset for improvement through subtle humor and empathy.

Stage

Iteration 5: developing the tool on a server to collect and build a library of seed/pre-seed pitches

Tools used

Animaze and Zoom for pitch obfuscation, Vibe-coded entirely using Claude Codeless than

14 of 30

Archive

Parsons E-LAB 2025

(10/10)

15 of 30

Phase 1

Early testing ways to achieve goals

APPROACH

16 of 30

INSTRUCTIONS for Recording Founders' Pitches on local computer for Elab Research Project Library

(Version of this for when we have a server to be updated)

Instructions:

1 .Open the Elab Zoom account (Critical) on the computer that has the anonymized program (Dell) only (make sure no other laptop is on the Zoom call to create an echo or noise)

2. Make sure that when you open a Zoom session, the screen name is “Parsons Elab” or “Research Project” and not a person’s name for their video screen name and no pronouns!!

NOTE:

Before each session, do a test run, of EVERY STEP, before each session, to make sure the system and protocol is working and we can capture the live recordings!!

FIRST RECORDING (with Avatar)

  1. Open Zoom, full screen, select the Avatar (with the voice mod), go full screen, and prepare to record
  2. Check that the screen name is “Parsons Elab” or “Research Project.”
  3. Record to the cloud with Avatar: press record, then stop recording immediately. Try not to have lagtime before or after, so no editing is needed. (Take over instead of edit for ost savings.)
  4. Save to cloud, when recording is ready>
  5. Download locally off the Zoom cloud (They automatically get deleted otherwis and upload to store in our project library in Google Drive (see link below)LIBRARY OF PITCHES

17 of 30

Previous Iterations

Zapier + LLM + ElevenLabs (Audio Only)

Using Zapier, we created a simple workflow wherein a file would be uploaded in the Drive folder, sent to ChatGPT/Gemini/Claude for anonymization processing and then would be vocalized using ElevenLabs before saving the output in the Drive folder once again

(07/10)

Prototype 1

PARSONS E-LAB

18 of 30

For prototype 2, we created a minimal workable prototype using Zoom Avatars and Voicemod to neutralize the visual identity and voice of the user.

(08/10)

Previous Iterations

Prototype 2

PARSONS E-LAB

For prototype 3, we added Animaze, an avatar animation platform, and imported avatars created in Viverse. Then combined the Animaze avatars, for visual identity masking, and Voicemod for voice modification with Zoom as the foundational application.

Previous Iterations

Prototype 3

19 of 30

(10/10)

Demo

Video conferencing platform for pitch delivery

Voice modulation software to mask accent and gendered vocal characteristics

Desktop application providing reactive avatars with upper and lower body tracking

Zoom

Voicemod

Animaze

Prototype 4

PARSONS E-LAB

Prototype 4

Claude AI Interface

20 of 30

SECOND RECORDING (Native)

  • Open Zoom on a computer, full screen, prepare to record >Record to the cloud
  • Press record (leave the cursor on the record stop start button in case the founder wants to stop and redo > stop recording right after. Try not to have time before or after, so there isn't any editing. Take over vs edit.
  • Save to cloud when recording is ready>
  • Download and store in our project library LIBRARY OF PITCHES

21 of 30

Workshop (Proof of Concept)

(09/10)

In addition to creating minimal viable prototypes (MVPs), we also organized a workshop where we invited VCs and a founder to participate in our research experiment.

PARSONS E-LAB

22 of 30

We used a claude generated AI model to simulate the AI platform we envision integrating into our platform and took a survey afterwards for feedback.

Brofolio AI

(08/10)

PARSONS E-LAB

23 of 30

General

Investment Decision

Founder Assessment

Risk Assessment

Open-ended Reflection

Sample Questions

  1. How likely are you to invest in this startup?
  2. Would you invest in this startup?
  3. The founder appears confident
  4. The founder appears resilient enough to handle startup challenges
  5. The founder appears capable of leading a growing team
  6. Market risk (market size, demand)

Link to Questionnaire

Defining

Questions for Workshop Participants

(08/10)

PARSONS E-LAB

24 of 30

(10/10)

Demo

1. Initialize Zoom

  • Log into Zoom with "E-Lab" Host account

2. Launch Avatar (Animaze)

  • Open Animaze application
  • Select session avatar
  • Enable video capture for face tracking

3. Configure Voice (Voicemod)

  • Open Voicemod application
  • Select voice profile
  • Enable microphone and test audio quality
  • Disable "Hear Yourself" after testing

4. Connect to Zoom

  • Set microphone to "Voicemod Virtual Audio Device"
  • Set camera to "Animaze"

Quick Setup Guide for recording pitches with/without obfuscation

PARSONS E-LAB

25 of 30

Workshop Setup

(09/10)

PARSONS E-LAB

we set up two rooms for the workshop. One for the experimental pitch with the AI bias detecting System and another as a control.

26 of 30

SYSTEM DIAGRAM

The the diagram below describes the version of the system we are currently working on, a library of recordings with standard questionnaires and an open AI to prompt with questions and feedback. It show the step where founders do anti-bias training, recording of anonymized and non-anonymized pitch recordings.

It also shows the step where the venture capitalists respond to questionnaires, ask questions and the trained AI responds with improvements and suggestions to be less biased. It also enlightens them on how they might gain more alpha if they stayed open to the project or took it to the next level and brought in the founder, how they get scored and how they come back for more training to improve their score. It shows how an office (VC Firm) or MBA program in an institution might have this system or portal set up and their goal to improve alpha and diversity of portfolio. It also shows our back end capturing data on learning and how this might translate into bigger returns for the investor or allocator.

27 of 30

Our Solution

(06/10)

PARSONS E-LAB

We address a persistent market inefficiency in seed-stage investing. The goal of this (anti-bias) training tool for capital allocators (early-career VCs, angel investors, and post-secondary MBA program students) is to help train investors to improve their portfolio performance by increasing their bias awareness when evaluating startup pitches, at the earliest funding stages.

WHAT

28 of 30

The open-source AI enabled platform would be accessible online. Users would log in to the platform and choose from a library of pre-recorded 5min videos seed-stage pitches from actuarial founders.

Some of the videos recordings are of mediated pitches where the founder uses a masked avatars with voice modulation and some are native. Investors hear the pitch, then they answer a questionnaire before going to an AI evaluation prompt/response area to “dialog” with our AI and getting a scorecard and can try again with other recordings in the library to learn to check their biases to improve their allocation and diversify their portfolio for bigger returns..

This isn't about hiding information; it's about reordering it.

After investors express genuine interest based on business merit, and some personality traits of the presenter, identities are revealed, and traditional due diligence begins. We're not replacing the investment process; we're making the first step fairer.

(06/10)

PARSONS E-LAB

We address a persistent market inefficiency in seed-stage investing. The goal of this (anti-bias) training tool for capital allocators (early-career VCs, angel investors, and post-secondary MBA program students) is to help train investors to improve their portfolio performance by increasing their bias awareness when evaluating startup pitches, at the earliest funding stages.

HOW

29 of 30

What is Beat The Odds (BTO) ?

(03/10)

A blind pitching tool where investors evaluate startups based purely on business fundamentals—no faces, no voices, no bias

PARSONS E-LAB

30 of 30

DIAGRAM DEPICTS: change to the latest version of the platform with the server, no longer live, but a library of recordings with a standard questionnaire and an open AI to prompt with questions and feedback. so a diagram with a person sitting in front of a computer alone doing anti-bias training. with an arrow to a cloud with a library of anonymized pitches by sector of pre-seed and seed pitches, and a library of live people pitching where we see real faces and not avatars.

show the step where the respond to questionnaire and ask questions> AI responds with improvements and suggestions to be less bias and how they might gain more alpha if they stayed open to the project or took it to the next level and brought in the founder, how they get scored and how they come back for more training to improve their score… shoe how an office (VC Firm) or MBA program in an institution might have this system or portal set up and their goal to improve alpha and diversity of portfolio…show then our back end capturing data on learning and how this might translate into bigger returns for the investor or allocator.