Inspiration
Small businesses apply to many different opportunities, from pitch competitions and buyer programs to grants, trade shows, and innovation programs.
But the preparation process is repetitive. Every time, founders have to read a new set of requirements, search through existing company materials, figure out what is already covered, identify what is missing, rewrite answers, and repeatedly ask AI tools to review the application.
I wanted to turn that fragmented process into a single readiness workflow that helps businesses reuse what they already know before asking them to provide more.
What it does
Opportunity Readiness Agent helps small businesses prepare for new applications and business opportunities by comparing what they already have against what the opportunity requires.
The agent:
- extracts and evaluates the required criteria
- checks those criteria against the business information already provided
- reuses relevant evidence where possible
- improves the draft using existing information
- avoids inventing missing business facts
- surfaces all unresolved required information together
- lets the user provide missing facts in one pass
- re-checks the full opportunity after new information is added
- produces an updated, review-ready draft
- generates a downloadable readiness report
Readiness is based on required-criteria coverage, rather than an arbitrary AI confidence score.
How we built it
The application is built in Python with Streamlit as the web interface and deployed as a public service on Google Cloud Run.
The agent uses the Google GenAI SDK (google-genai) to access Gemini 3.5 Flash through Vertex AI. Gemini is used to interpret opportunity requirements, evaluate existing evidence, identify gaps, and generate improved application content.
Firestore persists readiness audits, requirement status, evidence used, drafts, and workflow state.
The core workflow is:
User input → requirement audit → evidence reuse → gap detection → human input when needed → full re-check → final draft → downloadable readiness report
The system architecture is:
User → Streamlit UI → Google Cloud Run → Google GenAI SDK → Gemini via Vertex AI → Firestore
Challenges we ran into
One of the biggest challenges was deciding what the agent should do autonomously and what should require human input.
An early version asked users for missing information too frequently, which made the experience feel like another chatbot. We changed the workflow so the agent first reuses everything it safely can, then surfaces all remaining required gaps together.
Another challenge was readiness scoring. AI-generated percentage scores could vary between runs, so we moved the scoring logic out of the model and calculate readiness based on required-criteria coverage.
We also had to define clear trust boundaries. The agent should help rewrite and organize information, but it should not fabricate business metrics, production capacity, certifications, or competitive claims simply to make an application look complete.
Accomplishments that we're proud of
We built an end-to-end agent workflow that goes beyond a single prompt-and-response interaction.
The agent can audit an opportunity, reuse existing evidence, distinguish between information it can safely improve and facts that require human input, collect the unresolved gaps, and then re-check the full application after the user responds.
We also deployed the working application publicly on Google Cloud Run, connected Gemini through Vertex AI using the Google GenAI SDK, persisted workflow data in Firestore, and added downloadable readiness reports.
Most importantly, the workflow is designed around a real SMB problem: reducing the repeated preparation work that happens every time a new opportunity appears.
What we learned
The most useful agent experience is not simply generating more text.
A stronger workflow comes from deciding what can be completed automatically, what existing evidence can be reused, what should never be invented, and when human judgment is actually necessary.
We also learned how to build and deploy a stateful AI workflow on Google Cloud, use Gemini through Vertex AI with the Google GenAI SDK, persist agent state in Firestore, and design a human-in-the-loop workflow around real business preparation tasks.
What's next for Opportunity Readiness Agent
The current MVP starts with users pasting their existing business information and the requirements of a new opportunity.
A natural next step is URL ingestion, so users could provide an opportunity link and the agent could extract requirements, deadlines, requested materials, and evaluation criteria automatically.
The product could also evolve into a reusable business knowledge layer, allowing companies to maintain verified information once and reuse it across grants, buyer applications, trade shows, pitch competitions, partnerships, and other recurring opportunities.
Over time, the agent could help businesses track multiple opportunities, prioritize what is worth pursuing, and maintain an ongoing readiness profile instead of rebuilding every application from scratch.
Log in or sign up for Devpost to join the conversation.