AI Procurement Intelligence Pipeline

The Problem

For a small consultancy competing in UK public procurement, every tender bid represents a significant time investment before a single word of the response is written. Tracking opportunities across Find a Tender and Contracts Finder, reading dozens of lengthy specifications, assessing fit, and deciding whether to bid consumed hours the team didn't have -- and most of that effort was spent on tenders that were never a strong fit.

The Solution

An automated LLM pipeline that does the reading, scoring, and first-draft writing: opportunities are ingested from both procurement portals on a schedule, assessed against a structured capability profile using Claude, and those above the threshold are moved forward to bid-pack generation. The team only touches work that's already been pre-qualified as winnable.

How It Works

The pipeline runs on a cron schedule: pull new opportunities, parse and normalise the specification, score against the client's capability profile (sector fit, value range, submission window, past-performance signals), and write a structured assessment. For opportunities above the bid threshold, the system generates a draft bid structure -- section headings, compliance requirements, suggested evidence points -- ready for a human writer to complete. Decisions that previously took days now take minutes.

Technology

Built on Django with Celery for scheduling and task management. Claude (Anthropic) powers the scoring and generation layer. The scoring model is prompt-engineered with the client's specific capability profile and weighted evaluation criteria, making it tunable without retraining. Outputs are stored in the Django database and surfaced via a lightweight internal interface.