Svod RAG HR Copilot
Built a RAG-powered resume search system that lets HR professionals find candidates using natural language queries, with source attribution and instant results.
Svod — An HR tech solution designed to streamline resume screening and candidate discovery for Russian HR managers and recruiters.
Introduction & Background
HR professionals spend countless hours sifting through resumes, often struggling to find candidates that match specific, nuanced requirements. Traditional keyword search fails when recruiters want to find "someone with startup experience who has managed cross-functional teams." Svod needed a smarter way to search resumes—one that understood context and intent, not just keywords.
- Natural language queries replace complex boolean search strings
- RAG architecture ensures results are grounded in actual resume content
Problem & Constraints
The client needed a resume search system that could understand queries like "experienced Python developer who has worked in fintech" and return relevant candidates. The solution had to be affordable, easy to use for non-technical HR staff, and provide transparency by showing which parts of a resume matched the query.
- No dedicated infrastructure—needed managed cloud deployment
- Source attribution required for trust and verification
Solution Overview
We built a RAG (Retrieval-Augmented Generation) system that indexes resumes into ChromaDB with semantic embeddings. HR professionals can ask questions in plain Russian and receive ranked candidate matches with highlighted relevant sections from each resume.
- LangChain orchestration handles query understanding and retrieval
- ChromaDB provides fast vector similarity search over resume embeddings
Implementation Journey
The system parses uploaded resumes (PDF, DOCX), chunks them intelligently, and stores embeddings in ChromaDB. When a user submits a natural language query, LangChain retrieves relevant chunks and presents them with the original resume context. FastAPI provides a clean REST interface for the frontend.
- Resume parsing with robust format handling
- Semantic chunking preserves context boundaries
- OpenAI embeddings capture meaning beyond keywords
- Source attribution shows exactly which resume sections matched
Results & Impact
The delivered system allows HR professionals to search resumes using natural language, dramatically reducing time-to-candidate-discovery. The live demo showcases the core functionality and serves as a starting point for future enhancements.
- Affordable solution delivered via Kwork marketplace
- Scalable architecture ready for production HR workflows
Three phases, one acceptance test.
Requirements & design
Clarified resume parsing needs, search intent patterns, and ChromaDB integration requirements with the client.
Core development
Built FastAPI backend with LangChain RAG pipeline, resume ingestion, and natural language query interface.
Polish & deployment
Added source attribution, refined search relevance, deployed to Render with public demo access.
What it runs on.
The ones we came in with.
- MetricsWhat is the average time saved per candidate search compared to manual review?
- ScaleHow many resumes can the system handle before performance degrades?
- FeaturesWould multi-language support (English resumes) add value for the target market?
- IntegrationAre there ATS (Applicant Tracking System) integrations the client would prioritize?
Building something in this shape? Write us — fixed scope, fixed price, live in 30 days.