Skip to content
Page 5 of 6

Challenge 3: Context-Powered Iteration

Recap:

In Challenge 2, you turned your app from a prototype into a product. Hardcoded sample data became real data from live sources. You added pages, navigation, and polish, and everything was saved and synced so your whole team could see it.

Then in Lesson 3, you hit the wall: every new conversation meant re-explaining your entire project from scratch. You learned why AI forgets between conversations, and you solved it. Your team wrote a house-sitter note, turned it into a real project context file, and tested it. AI knew your project without being told.

You also practiced two ideas that matter right now:

  • Table of contents, not the whole book. Your context file is short and points to your project's documentation. It doesn't try to contain everything.
  • Start fresh for fresh eyes. When a conversation gets long, start a new one. With the context file in place, starting fresh costs you nothing. AI picks up right where you left off.

The Challenge

Your AI coding assistant knows your project now. Every conversation starts with context instead of confusion. That changes everything about how fast you can build.

In Challenge 1, you were learning how to talk to AI. In Challenge 2, you were learning how to work in a real project. In Challenge 3, you are just building, and context is the reason you can move this fast.

Time to be ambitious. Your Federal Vendor Intelligence Tool already shows what a vendor has done. Now add the layers that tell you whether you should buy from them and what risks to flag. SAM.gov registration verification, exclusion checks, socioeconomic certifications that matter for set-aside strategy, cross-referencing data across systems to catch mismatches, and data visualizations that surface patterns at a glance. The exact direction is up to your team. Context makes iteration fast enough to go wherever your ideas take you.

Build incrementally, one feature at a time, verified before you move on.

What to Build

Items are listed in priority order. If time is tight, focus on the items near the top first.

  • Your project context file is working. Start a fresh conversation and verify that AI knows your project, your data sources, and your design without you explaining anything
  • Exclusion check is part of the vendor profile: check the vendor against the SAM exclusions database and display a clear status indicator (green "Clear" or red "EXCLUDED"). This is the single most critical due diligence check (your repository already has exclusion data; ask your AI coding assistant what exclusion information is available)
  • SAM registration status: show whether the vendor has an active SAM registration, their UEI, and registration expiration date
  • Socioeconomic certifications: show 8(a), HUBZone, SDVOSB, WOSB status from SAM entity data, with badge-style indicators for each designation. Note that small business size standards are tied to specific NAICS codes; a company can be small under one NAICS code and large under another
  • Cross-reference check: compare the business_types from USAspending with businessTypeList from SAM to flag any mismatches between what a vendor claims in one system and what is registered in another. This is a core intelligence function: if the data does not agree, someone needs to investigate
  • Spending by agency visualization: a chart (bar chart, treemap, or whatever fits your tool's design) showing which agencies the vendor works with most. Acquisition professionals use this to assess past performance: Has the vendor done similar work, at similar scale, for agencies with a similar mission? A ranked list is not enough here; make it visual

These are options for teams that finish the baseline capabilities. You are the domain experts here. You do this work. Lean on that expertise to decide what to build next.

  • AI-powered vendor risk score: Define what "risk" means for a vendor evaluation based on your team's expertise. What factors matter? How would you weight them? Exclusion history, registration gaps, spending concentration in a single agency, business type mismatches, award volume trends? Design a risk model that reflects how you actually evaluate vendors, then encode that judgment into your tool. Have AI generate a weighted score with a plain-language explanation of each contributing factor. The goal is not a generic risk number; it is your risk framework, built into software.
  • AI-generated due diligence briefing: When a user views a vendor, generate a narrative intelligence report structured the way you would actually want to read one. Think about what goes into a real acquisition research memo: executive summary, past performance analysis, risk flags, contract vehicle pathways, competitive positioning, a recommendation, and gaps that require further investigation (CPARS ratings live in a restricted system; technical capability requires vendor demos or references). You know what a useful briefing looks like. Build one.
  • Competitive landscape intelligence: Given a vendor's NAICS codes, analyze all vendors in the dataset operating in the same space. Rank by award volume, identify market concentration, and generate a narrative. Is this vendor dominant, mid-tier, or a niche player? Are there enough qualified vendors to support competition or a set-aside? This is the kind of analysis that takes hours of manual research across multiple systems. Your tool should surface it in seconds.
  • Trend analysis with narrative: Go beyond a chart. Have AI interpret the spending patterns and generate a written analysis that tells the story behind the numbers. ("Award volume from Navy declined 40% between FY2022 and FY2023 while Air Force awards doubled, suggesting a shift in where this vendor is winning work.") The data is there. Make it talk.
  • Your call: What would you build that makes your job easier? What data do you wish was at your fingertips when you are researching a vendor? What takes you the longest today that software could accelerate? You have the data, you have the AI, and you have the domain expertise. Build the tool you wish you had.

About the SAM.gov data

Some features above use data from SAM.gov, the federal government's system for vendor registration and contract opportunities. Your project already has real SAM.gov data saved in the /data folder. Ask your AI assistant: "What SAM.gov data do we have in the data/beginner/ directory, and how can I use it?"

You do not need a SAM.gov API key. Everything in this challenge works with the pre-seeded data. If you want to try pulling live data as a stretch goal, you can request a free API key (requires a Login.gov account). Be warned: SAM.gov has strict daily request limits that you can hit fast. Tell your AI assistant to save every response so the same request is never made twice.

Tips

  • Start a fresh conversation first. Before you build anything, open a new conversation and make sure your context file is doing its job. Ask your AI coding assistant: "What is this project and what data do we have available?" If it knows, you are good to go. If it does not, fix the context file first. Everything else in this challenge goes faster when context is working.
  • Update your context file as you build. Tell your AI assistant to update the project context file when you: add a new data source (like SAM entities, exclusions, or opportunities), add a new page or section, change how the app is organized, or make a design decision you want AI to follow consistently (like "always show UEI as the primary identifier, not DUNS"). A quick "Update the context file to include what we just built" takes seconds and keeps future conversations sharp.
  • Try the discovery prompt. When you are ready to add new data sources, try: "What data do we have available in the data/beginner/ directory? Show me what's there and suggest how we could add it to the vendor tool." Let AI explore for you.
  • Save and sync often. You know the drill. This is your safety net. Save after every feature that works.

Go build. That is the brief. Spend the rest of this session block working on your challenge with your team. Your Facilitator will let you know when it is time for the Reflection.

Our recommendation: mob until you have clear swim lanes. Think of your project like a shared Google Doc. If two people write the same sentence in the same paragraph at the same time, their changes collide and one person's work gets overwritten. Code works the same way: when two people edit the same file at once, you get code collisions that lead to hard-to-trace bugs, lost work, or code that has to be thrown away. The safe way to split up is clear swim lanes, where each person works on a different section. Until your team has those swim lanes, mob: one person drives (hands on the keyboard), everyone else navigates (watches, thinks ahead, gives feedback). Rotate the driver every five minutes or so. One keyboard, zero collisions.