Explore Abel Adamu Buba’s AI engineering projects, their problems, architecture and outcomes.
Clinicians spent minutes hunting through PDFs, guidelines, and internal SOPs during patient encounters.
Retrieval-first agent over versioned clinical corpus with hybrid search, citations, and a strict refusal policy for out-of-scope queries.
Median lookup time reduced from minutes to seconds with citation-grade answers reviewers can trust.
High call volume and after-hours intake overwhelmed a small operations team, causing missed bookings.
Realtime voice agent with typed tool calls into scheduling, CRM, and payment APIs, plus a supervised handoff to humans on low-confidence turns.
Automated the majority of routine intake calls end to end with escalation paths that operators trust.
Manual extraction from mixed-format contracts and invoices was slow, inconsistent, and hard to audit.
Structured-output pipeline: OCR → layout parse → typed field extraction → validators → human-in-the-loop queue for anything below confidence.
Turned a multi-day queue into a same-day workflow with every field traceable to its source region.