What I built
A local Python and SQLite tool that pulls postings from public sources, scores each lead against a configurable profile, and uses an LLM review pass on the strongest candidates to catch fit and risk that keyword rules miss.
AI workflow tool
Contract Hunter
A local AI-assisted workflow that turns public job and contract postings into a ranked, human-reviewed queue, then prepares tailored application materials for approval.
From scattered listings to a working queue.
Contract and freelance leads arrive from job boards, contract feeds, and company career pages. Most are irrelevant, many are duplicates, and the useful work is deciding which ones deserve a tailored response.
Where it applies
The same approach can help with other lead-heavy work: sales and BD prospecting, recruiting pipelines, grant or RFP scanning, and partnership sourcing.
Architecture
Deterministic rules handle the broad pass first. LLM review is reserved for the candidates most likely to matter.
1Ingestion
Job boards, RSS, contract feeds, Greenhouse, Lever, and Ashby.
2Store
Normalized leads, feedback, notes, and draft status in SQLite.
3Scoring
Configurable weights, hard rejects, recency decay, and quick-project signals.
4LLM triage
Semantic category, fit reasons, risks, and score adjustment on top candidates.
5Review
Human decisions keep, set aside, or archive leads without deleting history.
6Draft
Tailored briefs, resumes, and application packets follow voice rules and require human approval.
Python
SQLite
LLM-assisted review
HTML/CSS/JS
YAML config
RSS + JSON feeds
ATS APIs
Engineering decisions
Deterministic rules score the full set, then LLM review runs where judgment is worth the cost.
Titles, skills, AI signals, hard rejects, and recency decay live in YAML.
The tool prepares the work for review. Nothing is sent automatically.
Drafts follow explicit anti-template rules, pricing guidance, and lead-specific context.
Redacted dashboard screenshots
These are captured from the working local tool with sample data in place of real leads. Private company names, outreach text, pricing, and notes are not published.
After a lead is selected
Build the application materials the opportunity actually asks for.
The editable outreach draft is one possible output, not the whole system.
Application noteA concise message shaped around the role and company.
Tailored resumeThe most relevant experience and proof moved forward.
Project examplesSelected work with links and role-specific context.
Complete packetThe requested pieces assembled into one coherent submission.
Human gate: review, edit, and approve every application before anything is sent.
Why this is transferable
I built Contract Hunter for my own search, then separated the parts that were specific to me from the general pattern: collect messy inputs, rank them, review the short list, and keep the final decision human.
Postings ingested and deduplicated across multiple public sources.
Cost-aware ranking: deterministic pass first, semantic fit checks second.
Review decisions and draft approval stay under human control.
Opportunity Radar, SignalFoundry, and Contract Hunter all use that pattern. The inputs change, but the judgment stays with a person.
Work together
I build tools like this.
Need a review queue, lead triage system, or AI-assisted drafting workflow with a human approval gate?