JeezAI

Overview

JeezAI was a solo attempt to lower the barrier to AI career access. Started in early 2023, it was an AI-native education platform that matched students with AI internships using autonomous agents. The goal was hands-on learning and community over gatekeeping.

The platform shipped an AI-vetted internship board, personalized career tools, and a community of US and Canadian students. I paused development to join 99Ravens. The multi-agent architecture and evaluation frameworks here fed directly into Expert AI work at 99Ravens.

Core Components:

• AI internships board. Autonomous agents scraped, researched, and curated internship opportunities. Multi-agent LLM systems handled search, relevance ranking, and summarization from companies like Hugging Face, Scale AI, and Astranis.
• DSPy matchmaking system. Cohere Command R+, Stanford NLP’s DSPy, Weaviate, and CrewAI agents. Analyzed student resumes with hybrid vector search and produced personalized matches.
• AI career tools. Cover letter generator, resume analyzer, LinkedIn profile analyzer, and a month-by-month career roadmap planner with book recommendations and leaders to follow.
• Hired in AI newsletter. Weekly newsletter featuring 5 internships, 5 AI/ML news items, and 5 startups.
• JeezAI AI Club. A community of university students for networking and peer learning.

Architecture

The most technically ambitious component was the DSPy Matchmaking System, which I open-sourced as a reference implementation for structured AI programming.

Multi-Agent Scraping Pipeline
Internship discovery ran on a CrewAI orchestration:

• Scraper Agent: Used Firecrawl and Browser Use to bypass anti-scraping measures on career pages
• Research Agent: Extracted company context, funding status, and tech stack from LinkedIn and Crunchbase
• Summarizer Agent: Generated concise internship descriptions with key requirements and deadlines
• Qualifier Agent: Filtered listings for genuine AI/ML roles vs generic software positions

All agents communicated via a shared state store, with human-in-the-loop approval for edge cases.

DSPy Compiler Framework
The matchmaking core used Stanford NLP’s DSPy for declarative structured programming:

# DSPy Signature for query generation
class generate_query(dspy.Signature):
    """Generate targeted search queries from resume analysis"""
    resume_text = dspy.InputField()
    career_goals = dspy.InputField()
    search_query = dspy.OutputField(desc="Optimized Weaviate query string")

# DSPy Module for internship matching  
class Internship_finder(dspy.Module):
    def __init__(self):
        self.generate_query = dspy.ChainOfThought(generate_query)
        self.hybrid_search = WeaviateRM("Internship", weaviate_client)
        self.analyze_match = dspy.Predict(generate_analysis)

DSPy’s compiler treated prompt engineering as an optimization problem, automatically improving query generation based on user feedback loops.

Vector Search & Hybrid Retrieval
Weaviate powered the internship database with Hybrid Search combining BM25 keyword matching and semantic vector search. This matched students based on both explicit skills (Python, PyTorch) and conceptual alignment (computer vision, NLP).

Cohere Command R+
Used for deep linguistic analysis of resumes and internship descriptions, generating structured JSON matches with confidence scores and mismatch explanations.

No-Code Infrastructure
As a solopreneur, I moved fast with Webflow for the frontend and Zapier for workflow automation between agents, Airtable for internship database, and Streamlit for the DSPy demo interface.

Status

Paused Q4 2024 to take the AI Agent Systems Manager role at 99Ravens. The multi-agent orchestration patterns, evaluation frameworks, and context qualification logic from JeezAI translated directly into the Expert AI architectures.

The project was useful as a proof that autonomous agents can solve real talent-matching problems at scale, and informed how I later thought about scaling strategic marketing expertise at 99Ravens.