BreezeRFP

Small and mid-size government contractors lose before they start: SAM.gov posts thousands of solicitations a week, most are a poor fit, and the ones worth chasing demand a proposal, a cover letter, and often a FOIA request — each written against a dense RFP PDF by a team that is already stretched. Capture teams either skim and miss opportunities or burn days responding to bids they were never going to win.

BreezeRFP splits the job into ingest, triage, and drafting. Celery workers poll the SAM.gov API on a schedule, normalise every solicitation into PostgreSQL, and an LLM matching pass scores each one against the company's profile so the feed is ranked by winnability rather than recency. For the response, each customer's past proposals and capability statements are chunked and embedded into a per-tenant FAISS index; LangChain retrieval grounds Claude 3.5 Sonnet and GPT-4o in that material so outlines, cover letters, and FOIA requests are written in the company's own voice with their own past performance. A smart reader distils long RFP PDFs into requirements and deadlines. The whole thing ships as a Vue 3 + Vuex SPA over a Django + DRF API, with Redis as the broker, S3 for documents, Stripe for billing, and Heroku for hosting.

• Celery workers ingest solicitations from the SAM.gov API and rank them by winnability with an LLM matching algorithm
• LangChain + FAISS RAG over each company's knowledge base grounds the AI assistant in their own past work
• Claude 3.5 Sonnet and GPT-4o draft proposal outlines, cover letters, and FOIA requests, plus a smart reader that distills long RFP PDFs
• Django + DRF API with PostgreSQL, Redis, and Stripe billing — deployed on Heroku with document storage on AWS S3