Building a SAM.gov Contract Scanner to Find Federal Work at Scale

The hardest part of being a newly certified small business in the federal market is not the registrations and it is not the certification, it is the simple problem of finding the work in the first place. The government posts thousands of solicitations across hundreds of agencies, the volume is genuinely enormous, and a one-person shop cannot read all of it by hand without burning every working hour on a search instead of on engineering. So we did the obvious thing for a software company and built a tool, an internal contract scanner that reads the federal opportunity feed for us, scores what it finds against what we can actually do, and tells us where to point our limited time. This is an honest writeup of how it works, because the tool itself is a good illustration of the differentiator we keep talking about, which is that AI-accelerated delivery is not a slogan for us, it is how a firm of our size competes against firms many times larger.

The problem the tool exists to solve

Heinrichs Software Solutions Company is an SBA-certified Service-Disabled Veteran-Owned Small Business, certified from June 4, 2026 through June 3, 2029, registered in SAM.gov with UEI SXG3SA9JMM47 and CAGE code 1ZSB5, carrying six NAICS codes with 541511 Custom Computer Programming Services as the primary. All of that is the price of admission, and none of it does anything on its own to put a relevant opportunity in front of us at the moment it is posted. The federal opportunity firehose is open to everyone equally, which means the advantage does not come from access to the data, it comes from how fast and how intelligently a firm can filter it. A large contractor solves this with a capture team of people whose entire job is to read solicitations and qualify them. We are one engineer and a Navy veteran founder, so we solved it the way engineers solve repetitive reading problems, by writing software that does the reading and leaves only the judgment to us.

Reading the SAM.gov Opportunities API across the matrix that matters to us

The core of the scanner calls the SAM.gov Opportunities API, the official public interface to the same solicitation data that contracting officers post, and it queries that feed deliberately rather than dumping everything. The query is built as a matrix, where one axis is our own set of NAICS codes and the other axis is the set of set-aside types that a firm like ours is eligible to pursue, which includes service-disabled veteran-owned set-asides, veteran-owned set-asides, total small business set-asides, and the unrestricted competitions that anyone can bid. Crossing those two axes means the tool is not asking the broad question of what is the government buying, it is asking the precise question of what is the government buying inside the lanes where our certification and our skills actually apply. That framing alone removes an enormous amount of noise before any scoring happens, because the great majority of federal solicitations are simply not in a NAICS code we carry or a set-aside we qualify for.

On top of that we drop entire categories of work that are technically adjacent in the classification system but have nothing to do with what a software and cybersecurity firm delivers, so construction, janitorial services, and the other facilities and labor categories that surface around IT codes get filtered out before they ever reach a human. The point of the matrix and the category drops together is to make sure that the opportunities that survive the first pass are at least plausibly things we could actually do and actually win, rather than a long list that someone still has to manually disqualify one row at a time.

Scoring every opportunity from zero to one hundred

Filtering tells us what is eligible, but eligibility and fit are not the same thing, so the scanner assigns each surviving opportunity a score from zero to one hundred that estimates how good a match it really is for us. The score is a weighted blend of a few honest signals. The first is set-aside match, because an SDVOSB set-aside where we compete only against other certified veteran-owned firms is structurally more winnable for us than an unrestricted competition against the open field, and the scoring reflects that difference in standing. The second is skill-keyword density in the title and the description, which is a direct measure of how close the actual scope of work sits to the engineering we do, since a solicitation that repeatedly names the technologies and capabilities in our portfolio is a far better fit than one that merely shares a NAICS code with our work. The third is the estimated value of the opportunity, which helps us weigh whether a given pursuit is worth the time it will cost to chase. The fourth is the number of days remaining to respond, because an opportunity closing in two days is a very different proposition from one with a month of runway, and the time pressure has to be part of how we rank the queue.

Blending those signals into a single number is not magic and we do not pretend it is, it is a transparent weighting that turns a pile of eligible solicitations into a ranked list where the top of the list is genuinely worth a human reading in full. The score does not decide anything by itself, it decides what we read first, and for a firm with a finite number of hours that ordering is most of the value.

Turning awards into a teaming target list

A bid list is useful, but it is only half of how a new firm without past performance actually breaks in, and the more important half is teaming. So the scanner also queries USAspending.gov, the public no-authentication API that publishes where federal money actually goes, and it pulls recent SDVOSB awards inside the same NAICS codes we carry. Rather than treating those awards as historical trivia, the tool aggregates them by recipient and ranks the prime contractors who have been winning the most relevant work in our lanes. The result is a target list of the firms most worth approaching about teaming and subcontracting, which is a completely different and arguably more valuable artifact than a list of open solicitations.

This matters because of the honest chicken-and-egg problem every new entrant faces, where the work tends to flow to firms that have already done the work, and a company with no federal past performance has to find another way onto a contract. Subcontracting to an established prime is one of the cleanest ways across that gap, since it lets a capable shop deliver real federal work and build a track record under the umbrella of a firm that already holds the relationship. By mining the award data for the primes who are demonstrably active in our exact NAICS codes and the SDVOSB space, the tool gives us a researched, ranked list of who to talk to, so our outreach is aimed at firms with a current appetite for exactly the kind of partner we are rather than scattered across names we picked at random.

How it runs and why it is built the way it is

The whole thing is packaged as a Docker container, which keeps the runtime and its dependencies reproducible and means the same tool behaves identically whether it runs on a laptop or on a schedule in the cloud. It persists everything it finds in a local SQLite database, which serves two purposes that both matter for a tool meant to run repeatedly. The first is deduplication, because the same solicitation appears across multiple runs and we do not want to surface an opportunity we already triaged as if it were new, so the database remembers what it has already seen and only flags what genuinely changed. The second is that keeping a durable record lets the firm look back across time at what has been posted in our lanes, which is its own kind of market intelligence for a company learning the rhythm of how its niches buy. On each run the scanner writes a dated markdown digest, a clean readable summary of the top-scored opportunities and the current teaming target list, which is the actual thing a human sits down and reads.

Because it lives in a container and produces a digest, it can run unattended on a daily schedule, which is the entire point. The federal market does not wait for anyone to have a free afternoon, opportunities open and close on their own clock, and a daily automated pass means we learn about a relevant solicitation while there is still real time to respond to it rather than discovering it after the window has closed. Choosing SQLite and markdown and a single container rather than a heavier stack was deliberate, because the tool only has to serve one firm well and the simplest design that does the job reliably is the right one to maintain when the engineer maintaining it is also the engineer doing the billable work.

Why this is the honest version of finding work at scale

We want to be clear about what this tool is and is not, because the federal space is full of overpromising and we would rather be trusted than impressive. It does not win contracts, it does not write proposals, and it does not replace the judgment that decides whether a given opportunity is truly worth pursuing. What it does is let one certified shop see the part of the federal market that is actually relevant to it, ranked by fit, refreshed every day, alongside a researched list of the primes worth teaming with, and that is a genuinely large multiplier on the time of a firm that does not have a capture team. We are newly certified and we have not yet won a federal contract, and we are not going to dress this up as anything other than what it is, which is the pipeline-building engine of a one-person shop that intends to compete on the strength of the work. Building it was a few days of engineering that pays itself back every single day it runs, and it is a fair illustration of the way we approach the whole business, which is to use software and automation to do at the scale of a small firm what would otherwise take a team. That is what AI-accelerated delivery means here, and a tool that quietly reads the federal market for us every morning is a good place to see it working.