Raah
I Built an AI System That Finds Real Business Problems and Turns Them Into Experience for Graduates
The Problem I Could Not Stop Thinking About
Every year thousands of graduates leave university with real skills and zero proof of them.
They apply for jobs. The job requires experience. They do not have experience. They cannot get experience without a job. Nobody breaks this loop.
I surveyed 15 graduates in my network. 80% said the experience trap was their biggest barrier. 93% said they would try a team-based real-work solution if one existed.
The existing solutions did not work. Job boards require experience you do not have. Internship platforms are competitive and scarce. Online courses give you certificates nobody trusts. LinkedIn is a platform for people who already have networks.
I kept thinking what if instead of connecting graduates to jobs, you gave them real work before anyone would hire them?
That question became Raah.
What Raah Does
Raah is not a job board. It is not a learning platform. It is not a freelance marketplace where graduates bid against professionals with five years of experience.
Raah is an AI agent system that finds real unsolved business problems posted publicly online, qualifies them through multi-step reasoning, and matches student teams to them automatically.
The graduate does nothing to find the work. The agent hunts it for them.
When a team completes a real project and the business confirms the delivery, a verified work record is created. Not a certificate. Not a course completion badge. Proof of real work the original business problem, the deliverable, the business owner’s confirmation. Something an employer can verify.
Why Existing Platforms Fail
Most platforms start by asking businesses to post opportunities. That creates a cold-start problem: businesses will not join until students exist, and students will not join until opportunities exist.
Raah solves this differently. Instead of waiting for businesses to come to the platform, the OpportunityHunter agent discovers real business problems already being discussed publicly online. Opportunities exist before the first business signs up, allowing the platform to create value from day one.
How I Built It
The Agent Pipeline
The core of Raah is three AI agents running in sequence using Google ADK 2.0.
Opportunity Hunter Agent uses Groq Llama 3.3 70B together with Exa and Tavily search APIs to discover real business problems posted publicly online. It searches founder communities, startup forums, Reddit discussions, Indie Hackers, LinkedIn posts, and other public sources. The agent filters results, removes duplicates, validates relevance, and returns structured opportunities with source URLs, dates, and business context.
Qualifier Agent uses Groq Llama 3.3 70B to reason through each post across four steps:
- Is this a real problem or just vague frustration?
- Is it specific enough that an outsider could help?
- Can it be solved without internal access or physical presence?
- Is there a clear deliverable achievable in one to two weeks?
Each post gets scored on specificity, solvability, and effort fit. Only posts scoring six or above on all three pass through. Everything else is rejected with a reason.
Brief Generator Agent takes every passed post and converts it into a structured task brief — problem statement, business context, deliverable, skills needed, timeline, and difficulty level. The original words from the business owner are preserved exactly.
Why Google ADK 2.0
Google ADK 2.0 provides workflow orchestration and shared state management across agents. OpportunityHunter, Qualifier, and BriefGenerator execute sequentially while sharing context automatically. I chose Groq Llama 3.3 70B for agent reasoning because of its speed, low latency, and strong reasoning performance. Exa and Tavily provide the retrieval layer, allowing the system to discover opportunities across multiple public sources without relying on a single search provider.
The Matching System
Get Farah Maqbool’s stories in your inbox
Join Medium for free to get updates from this writer.
After the pipeline saves qualified briefs to MongoDB Atlas, a separate matching script reads all open briefs and all registered users. It computes skill overlap between each user’s listed skills and each brief’s required skills. Teams of one to four people are formed automatically based on best match. Assignments are saved to MongoDB. Each person can only be in one active team at a time.
The User Experience
Graduates sign up through a Streamlit interface. They list their skills and field of study. When they log in they see only their assigned brief — not everyone else’s. They see their team members, the original business context is preserved through concise summaries while maintaining factual accuracy and source attribution..
They update their status as the project progresses discussed internally, pitched to the business, hired, work submitted. When they submit a final deliverable the verified record is generated.
Team communication happens over WhatsApp. The team leader creates a group and pastes the invite link into the platform. Everyone on the team sees a join button. No chat infrastructure to maintain.
What the Agent Actually Does A Real Example
Here is what happened when I ran the pipeline:
In one test run, the OpportunityHunter discovered a post from a SaaS founder struggling with B2B sales conversion. Their analytics tool solved a real problem but conversations kept dying before closing. They were getting into calls with decision-makers but could not create urgency.
The Qualifier reasoned through it:
Is this real? Yes specific struggle, specific product, specific symptom.
Specific enough? Yes SaaS tool for D2C brands on quick commerce platforms, defined sales problem.
Outsiders can solve it? Yes sales strategy research, value proposition analysis, stakeholder mapping all possible externally.
Bounded? Yes deliverable is a revised sales strategy and outreach framework, achievable in ten days.
Scores: Specificity 9, Solvability 9, Effort fit 9. Passed.
The BriefGenerator produced a clean task brief. A team was matched. A graduate with B2B sales skills and a graduate with market research skills were assigned together.
That graduate came to the platform with no work experience. They left with a verified record showing they developed a B2B sales strategy for a real SaaS product with the original Reddit post as proof the problem was real.
What I Learned Building This
The agent is the product. Everything else the UI, the database, the matching logic is infrastructure. The value is entirely in whether the agent finds real problems and qualifies them honestly. If it hallucinates or produces vague output the entire system loses credibility.
Prompt engineering matters more than model choice. The qualifier prompt went through many iterations. The most important addition was a single rule: if the previous agent returned nothing, say so and stop. Do not invent posts to fill the quota. That one line eliminated hallucination almost entirely.
Two-sided marketplace problems are real. You need businesses with real problems and graduates to solve them simultaneously. The agent finding problems from public forums rather than requiring businesses to post on the platform is what breaks the chicken-and-egg problem. Businesses do not need to know Raah exists for their problems to appear in the system.
The experience trap is a confidence problem as much as a structural one. From my survey research several graduates had not applied for many jobs not because opportunities did not exist but because rejection felt inevitable before they even tried. A platform that gives them a real win — even a small one — changes that.
What Is Next
Raah is working as a prototype. The agent pipeline finds real problems. Teams get matched. The UI shows assigned briefs. Status tracking works. WhatsApp groups connect teams.
What comes next is the real platform proper authentication, mobile-friendly interface, automated scheduling so the agent runs continuously, and the verified work record page that becomes a graduate’s proof of capability.
The goal is not to become another LinkedIn. The goal is to be the place graduates go before they are ready for LinkedIn where they get their first real proof of what they can do.
Try It
GitHub: https://github.com/Farah-Maqbool/raah
Built with Google ADK 2.0, Gemini 2.5 Flash, Groq Llama 3.3 70B, MongoDB Atlas, and Streamlit.
