AI-assisted recruitment screening
A recruiting platform with semantic resume search, JD matching, an emailed AI interview chatbot with explainable scorecards, and a Kanban pipeline that auto-advances candidates.
The challenge
Recruiters hold large resume pools they cannot search meaningfully and spend hours on first-round screening. Scoring is inconsistent and hard to explain. Hiring teams also want to reduce bias in early review while keeping candidate data separated per organisation.
What we built
A FastAPI service stores resumes in object storage and parses them with Gemini via Celery workers. It embeds them with Vertex AI into PostgreSQL with pgvector (HNSW index) for semantic and hybrid search, natural-language query parsing, and JD matching. A screening interview is emailed to candidates; Gemini generates questions, voice is synthesised, and answers are scored into scorecards that advance the pipeline. Every table carries a tenant id and all queries are scoped to the token's tenant. A React SPA provides the pipeline, dashboard, alerts and analytics.
Key capabilities
- Semantic and hybrid resume search with pgvector
- JD-to-candidate matching and ranking
- Emailed AI interview with screening and technical rounds
- Explainable scorecards and automatic pipeline stage advance
- Optional bias-blind review mode
- Daily attention alerts and screening analytics
What it delivers
- Resume pools become searchable by meaning rather than by keywords
- First-round screening is structured and scored consistently with explainable results
- Tenant scoping is enforced in queries and covered by a dedicated cross-tenant leakage test
- Uploads are de-duplicated and processed asynchronously so the UI stays responsive
Related work
Asset lifecycle and procurement system
An asset management platform covering lifecycle, maintenance, budgets, procurement workflows, leases, helpdesk and invoice OCR, with Gemini-assisted categorisation and analytics.
Read the case study →
Insurance claims assessment with deterministic gates
An agentic pipeline that classifies motor and medical-aid claim PDFs, reconciles every figure in plain code, and returns a structured review plus an annotated PDF.
Read the case study →
Insurance quote generation and policy comparison
A broker-facing platform that parses policy documents with an LLM, builds explainable premium quotes, versions them, and compares quotes side by side.
Read the case study →Have a similar backlog?
Most of our work started as a single painful workflow. Tell us yours.