Below is a direct copy-paste “About the Project” section for your Devpost portfolio, with a visually structured architecture and workflow that matches the actual AccessHire implementation.
🚀 About the Project
AccessHire — Adaptive Capability Twin
See Capability. Prove Potential. Enable Transition.
AccessHire is an AI-powered Career & Workforce Operating System that transforms real-world experiences into measurable, evidence-backed capabilities and uses those capabilities to support career mobility, practical verification, learning, and workforce decisions.
Instead of relying primarily on job titles, degrees, certifications, resume keywords, or career continuity, AccessHire focuses on what a person has actually done, demonstrated, and can prove.
🧠 The Core Idea
┌──────────────────────┐
│ REAL-WORLD │
│ EXPERIENCE │
│ │
│ Projects • Work │
│ Volunteering • Life │
│ Self-Learning │
└──────────┬───────────┘
│
▼
┌──────────────────────┐
│ CAPABILITY │
│ TRANSLATION │
│ │
│ Fine-Tuned MPNet ML │
│ + 469 Capabilities │
└──────────┬───────────┘
│
▼
┌──────────────────────┐
│ CAPABILITY TWIN │
│ │
│ Capability │
│ Confidence │
│ Evidence │
│ Context │
│ Verification │
└──────────┬───────────┘
│
▼
┌──────────────────────┐
│ PRACTICAL PROOF │
│ │
│ Micro-Task → Output │
│ → Evaluation │
└──────────┬───────────┘
│
▼
┌──────────────────────┐
│ CAREER TRANSITION │
│ │
│ Opportunities │
│ Learning Paths │
│ Capability Gaps │
└──────────────────────┘
🔥 What Makes AccessHire Different?
Traditional systems often follow:
Resume
↓
Keywords
↓
Job Match
↓
Candidate
AccessHire follows:
Experience
↓
Discover Capability
↓
Attach Evidence
↓
Identify Capability Gaps
↓
Practical Work
↓
Verify Capability
↓
Update Capability Twin
↓
Match Opportunity
↓
Enable Transition
The fundamental shift:
From predicting potential → to creating evidence that proves potential.
🧠 Custom ML Capability Engine
At the heart of AccessHire is a domain-adapted MPNet-based semantic capability inference model.
The model was adapted specifically for mapping human experiences to standardized workforce capabilities.
ML Pipeline
Experience Text
│
▼
┌──────────────────┐
│ Preprocessing │
│ & Segmentation │
└────────┬─────────┘
▼
┌──────────────────┐
│ Fine-Tuned MPNet │
└────────┬─────────┘
▼
┌──────────────────┐
│ 768-D Embedding │
└────────┬─────────┘
▼
┌────────────────────────┐
│ Capability Embeddings │
│ 469-Node Taxonomy │
└───────────┬────────────┘
▼
┌────────────────────────┐
│ Semantic Similarity │
└───────────┬────────────┘
▼
┌────────────────────────┐
│ Keyword + Evidence │
│ Signal Analysis │
└───────────┬────────────┘
▼
┌────────────────────────┐
│ Confidence Fusion │
└───────────┬────────────┘
▼
Ranked Capabilities
Example
Input:
“I organized my village's annual festival, managed vendors, coordinated volunteers, and controlled the budget.”
AI discovers:
Event Operations 88%
Vendor Management 84%
Budget Management 81%
Volunteer Coordination 79%
Resource Management 75%
The experience does not need to explicitly say:
“I have Vendor Management skills.”
The model identifies the capability from the activity itself.
🧪 Hard-Negative Learning
AccessHire doesn't rely only on obvious positive and negative examples.
During model adaptation, semantically similar but incorrect capabilities can be used as hard negatives.
Correct Capability
│
▼
Find Similar Capabilities
│
├── Related Capability A
├── Related Capability B
├── Related Capability C
└── Incorrect Capability
│
▼
Hard Negative
This helps the model learn the difference between capabilities that are semantically close but functionally different.
📊 Adaptive Capability Twin
Instead of storing:
Python = YES
AccessHire maintains a richer capability representation:
┌───────────────────────────────┐
│ CAPABILITY TWIN │
├───────────────────────────────┤
│ Capability │
│ Confidence │
│ Proficiency │
│ Evidence │
│ Context │
│ Recency │
│ Verification Status │
└───────────────────────────────┘
The Twin can evolve as new experiences, evidence, assessments, and practical work are added.
🧪 4-Step Practical Verification
AccessHire doesn't stop at inference.
┌───────────────┐
│ 1. MICRO-TASK │
└───────┬───────┘
↓
┌───────────────┐
│ 2. SUBMIT │
│ OUTPUT │
└───────┬───────┘
↓
┌───────────────┐
│ 3. AI │
│ EVALUATION │
└───────┬───────┘
↓
┌───────────────┐
│ 4. TWIN │
│ UPDATED │
└───────────────┘
This creates a feedback loop:
Infer → Test → Verify → Update
🤖 Multi-Agent Career Orchestration
AccessHire combines specialized AI agents into a coordinated workflow.
┌─────────────────────┐
│ CAREER ORCHESTRATOR│
└──────────┬──────────┘
│
┌──────────────┬─────────────┼──────────────┬──────────────┐
▼ ▼ ▼ ▼ ▼
Skills Market Learning Inclusive Bias
Discovery Intelligence Pathway Matching Audit
│ │ │ │ │
└──────────────┴─────────────┼──────────────┴──────────────┘
│
▼
Equity / Fairness
│
▼
Human Review
│
▼
Career Action
Specialized agents include:
- SD — Skills Discovery Agent
- MI — Market Intelligence Agent
- LP — Learning Pathway Agent
- IM — Inclusive Matching Agent
- BA — Bias Audit Agent
- EN — Equity Nudge Agent
- JF — Job Fairness Agent
- AA — Accessibility & Accommodation Agent
🎯 From Capability to Career Transition
Once capabilities are identified, AccessHire determines what the person can potentially transition into.
Current Capability Twin
│
▼
Target Opportunity
│
▼
Capability Gap
│
▼
Minimum Learning
│
▼
Practical Micro-Task
│
▼
Verification
│
▼
Updated Capability
│
▼
Career Transition
The goal is not to force a person through a generic course catalogue.
It is to identify the minimum practical path needed to demonstrate readiness.
⚖️ Responsible AI
AccessHire is designed around:
AI recommends. Humans decide.
The platform incorporates:
- Evidence-grounded capability scoring
- Explainable ML signals
- Human-in-the-loop safeguards
- Bias auditing
- Equity nudges
- No fabricated credentials or experiences
- Accessibility-aware interfaces
The system is designed as decision support, not an autonomous hiring authority.
🏗️ Technical Architecture
ACCESSHIRE
│
┌─────────────┴─────────────┐
│ │
Next.js Frontend AI Services
│ │
▼ ┌──────┴──────┐
Career / Workforce │ │
Interface ML Engine AI Agents
│ │
▼ ▼
FastAPI + MPNet Generative AI
│ │
└──────┬──────┘
▼
Capability Intelligence
│
┌──────────────────┼─────────────────┐
▼ ▼ ▼
Capability Twin Opportunities Verification
Technology Stack
| Layer | Technology |
|---|---|
| Frontend | Next.js 16, React 19, TypeScript |
| UI | Tailwind CSS, Framer Motion |
| ML Backend | FastAPI |
| ML Model | Fine-tuned MPNet Sentence Transformer |
| Semantic Search | Embedding + Cosine Similarity |
| Capability Taxonomy | 469 capabilities |
| AI Agents | Generative AI |
| Data | Evidence-backed capability profiles |
| Deployment | Netlify + FastAPI inference service |
👥 Real-World Inclusion Use Case
Consider a person returning to work after a career break.
Instead of:
3-Year Career Gap
↓
Lower Candidate Score
AccessHire can process the person's actual experiences:
Caregiving Coordination
+
Medical Scheduling
+
Budget Management
+
Vendor Negotiation
+
Self-Learned Python
↓
Transferable Capabilities
↓
Capability Verification
↓
AI Operations Opportunity
The system therefore changes the question from:
“Why was this person away from work?”
to:
“What capabilities did this person build, and can they prove them?”
🌟 Why We Built AccessHire
The future of work should not be determined solely by where someone worked, what degree they hold, or whether their career history is continuous.
People build valuable capabilities through projects, communities, volunteering, entrepreneurship, caregiving, self-learning, and lived experience.
AccessHire turns those experiences into structured capability evidence and creates a pathway from evidence to opportunity.
See Capability. Prove Potential. Enable Transition.
Built With
- ai-agents
- capability-mapping
- capability-twin
- cosine-similarity
- css
- fastapi
- framer-motion
- gemini
- generative-ai
- javascript
- machine-learning
- mpnet
- next.js
- python
- react
- semantic-embeddings
- sentence-transformers
- tailwind-css
- typescript
Log in or sign up for Devpost to join the conversation.