Inspiration
Small government contractors face thousands of federal notices, dense solicitation packages, and short response windows. Finding an opportunity is only the beginning: teams still need to understand the buyer, identify mandatory requirements, assess fit, and decide whether a pursuit deserves scarce proposal resources.
FedPulse was built to compress that workflow from hours of document review into one traceable decision experience.
What it does
FedPulse lets a supplier:
- Search and rank SAM.gov opportunities.
- Open an opportunity intelligence brief.
- Inspect submission, evaluation, socioeconomic, pricing, and compliance signals.
- Review source coverage and extraction confidence.
- Ask a GPT-5.6 procurement analyst to stress-test the pursuit.
- Receive a grounded recommendation and concrete next action.
The Decision System also exposes the broader Federal Procurement Intelligence Engine: semantic graph search, procurement Q&A, opportunity, supplier and department intelligence, similarity and recommendation engines, graph analytics, six strategy agents, compliance controls, proposal win themes, pricing risk, and a human-gated executive decision.
The demo uses a curated June 2026 SAM.gov snapshot. Every claim is bounded by source evidence, and users are reminded to verify the live notice before acting.
How we built it
A pre-existing research archive contained procurement-processing scripts, extracted requirement records, and proposed interface contracts. It did not contain a runnable web application, HTTP API, or OpenAI-powered user workflow.
During Build Week, Codex helped audit the archive, isolate defensible SAM.gov assets, design the product flow, create a new repository, implement the responsive application, add the server-side OpenAI Responses API integration, write validation tests, document the prior/new work boundary, and deploy the public demonstration.
GPT-5.6 is used at runtime as a bounded procurement analyst. It receives a structured evidence packet rather than unrestricted raw context, is instructed not to invent requirements, and produces a concise recommendation, rationale, and next action. Deterministic search and evidence retrieval remain separate from model reasoning.
Challenges
- Separating pre-existing research from Build Week work.
- Discovering that some production API assets were contracts and registries, not an implemented service.
- Reducing a broad intelligence platform into one coherent supplier workflow.
- Preserving provenance and uncertainty while keeping recommendations useful.
- Publishing a judge-friendly sample without exposing bulk datasets or credentials.
Accomplishments
- Built a complete runnable product experience rather than a dashboard mockup.
- Connected structured SAM.gov evidence to live GPT-5.6 reasoning.
- Made bid/no-bid decisions, evidence, risks, and next actions understandable in one flow.
- Added a transparent fallback so the demonstration remains testable during external service interruptions.
- Documented exactly what existed before Build Week and what Codex created during the competition window.
What we learned
Procurement teams do not need another generic chatbot. They need reliable retrieval, explicit evidence boundaries, and a model that reasons over controlled decision context. The strongest product experience starts with the decision a user must make—not with the size of the underlying dataset.
What's next
- Live SAM.gov synchronization and amendment monitoring.
- Supplier onboarding using capabilities, NAICS codes, certifications, and past performance.
- Evaluation-factor-to-win-theme mapping.
- Collaborative compliance matrices and proposal-volume ownership.
- Historical award and incumbent intelligence.
- Recommendation-quality evaluation with capture professionals.
Built With
- drizzle-orm
Log in or sign up for Devpost to join the conversation.