Problem
A long list looked productive but did not support a decision.
Sources used different formats for dates, budgets, eligibility, buyer proof, and risk. Weak matches and duplicates could demand the same attention as credible work.
Independent build · Private workflow
Turn scattered opportunities into decisions a human can defend.
Acquisition Autopilot collects and normalizes opportunity evidence, explains its ranking, and keeps every message or submission behind review.
Build
One record now carries the evidence and the reason.
I built canonical opportunity records, deterministic qualification rules, explainable ranking, private administrator routes, and review states for drafts and next actions.
Result
The queue says verify, act, or reject—and why.
Reviewers can compare fit and risk without losing source context. The workflow stops before outreach or submission, so ranking never becomes unsupervised external action.
Before / after
Put evidence before action.
Broken
decision = rank(source_item)
send(decision)Fixed
record = normalize(source_item)
signals = qualify(record.evidence, record.risk)
decision = rank(signals)
require_admin(request)
render_review(record, decision, signals.reasons)Evidence
The public proof shows the review surface, not a sales result.
- The diagram above is an architecture map of the implemented review boundaries; it contains no prospect or customer records.
- Repository checks cover authentication, scoring, imports, draft handling, private artifacts, responsive layout, and privacy controls.
- No test count is stated here because a fresh count was not run for this page.
Stack and capabilities
FastAPI workflow engineering with explicit control points.
- FastAPI
- SQLAlchemy
- Alembic
- SQLite
- Jinja
- Authentication
- CSRF protection
- Explainable scoring
Next