AI Hiring Platform

DeepScreen

DeepScreen is an end-to-end AI hiring platform that runs from resume ingestion to offer decision in a single integrated system. It replaces spreadsheets and manual handoffs with instrumented stages, logged decisions, and consistent evaluation. The platform combines a Rust backend, Tauri desktop proctoring, Qdrant vector search, and offline AI assessment to screen, shortlist, assess, and report on candidates at scale.

Type

AI Hiring Platform

Stack

9 Technologies

Features

6 Key Features

Status

Live

DeepScreen preview
◉01. Overview

DeepScreen is an end-to-end AI hiring platform that runs from resume ingestion to offer decision in a single integrated system. It replaces spreadsheets and manual handoffs with instrumented stages, logged decisions, and consistent evaluation. The platform combines a Rust backend, Tauri desktop proctoring, Qdrant vector search, and offline AI assessment to screen, shortlist, assess, and report on candidates at scale.

Built with Rust, Tauri, Qdrant, PostgreSQL, Groq, ONNX, LiveKit, Judge0, React. This project showcases end-to-end development from architecture to deployment, demonstrating expertise in modern web technologies and production-grade engineering.

◉02. Features
✓

End-to-End Pipeline

Runs from resume ingestion to offer decision in one system — every stage instrumented, every decision logged, every candidate evaluated consistently with no manual handoffs.

✓

AI-Powered Matching

Semantic resume analysis with section-wise embeddings and skill gap identification via vector arithmetic.

✓

Intelligent Interviews

Real-time skill assessment with adaptive questioning, offline TTS with 3D avatar lip-sync, and OS-level integrity verification via Tauri.

✓

Gated Assessments

Candidates must clear each stage before advancing — coding assessment, AI avatar interview, then live recruiter session — so expensive recruiter time is reserved for proven candidates.

✓

Configurable Filters

Semantic match score, education level, experience range, skill keywords, location preference, and recency filters, each adjustable per role.

✓

Hiring Analytics

Time-to-hire, cost-per-hire, pipeline velocity, and source attribution surfaced through recruiter, department, candidate funnel, and executive dashboards.

◉03. Functionalities
⚙

Resume Parsing

EdgeParse, a pure-Rust PDF-to-Markdown converter, extracts text in 0.064s per document with a 0.787 benchmark score and no JVM or GPU dependency.

⚙

LLM Structuring

Groq API runs a two-stage pipeline — structure_to_markdown normalizes into canonical sections, then extract_resume_fields produces typed JSON, with regex ground truth for links.

⚙

Semantic Embedding

Four separate all-mpnet-base-v2 ONNX INT8 embeddings per resume (Skills, Projects, Experience, Summary), outperforming whole-document embedding by 11.4pp on Precision@10.

⚙

Vector Storage

Embeddings stored across four Qdrant collections via gRPC (1.11ms p50 vs 2.40ms REST); PostgreSQL stores metadata, extracted links, and Cloudinary PDF URLs.

⚙

Shortlisting Engine

Three-stage retrieve-then-rerank — HNSW vector retrieval builds a 2K pool (0.20), cross-encoder re-ranking narrows to top-K (0.65), and BM25 lexical scoring recovers exact matches (0.15).

⚙

Coding Assessment

Judge0 CE sandbox with hidden test cases and time complexity analysis; Tauri desktop client monitors for AI copilots, remote desktop tools, and browser automation.

⚙

AI Avatar Interview

sherpa-onnx + Supertonic 3 TTS (99M params, 31 languages, CPU-only, <50ms init) with adaptive LLM questioning via Groq (350–600ms response).

⚙

Live Recruiter Session

Multi-participant WebRTC via LiveKit SFU — 80–150ms audio, 120–250ms video, <300ms multi-user sync, <15% CPU with 4 participants.

⚙

Proctoring Signals

MediaPipe Face Mesh gaze tracking, solvePnP head pose, face presence detection, YOLOv8s multi-person and prohibited object detection, and Tauri sysinfo process blocklist monitoring.

⚙

Composite Reporting

All assessment signals fused into a configurable weighted score, delivered as an evidence pack with resume summary, coding scores, interview transcript, proctoring logs, and integrity confidence.

◉04. Tech Stack
Rust
Tauri
Qdrant
PostgreSQL
Groq
ONNX
LiveKit
Judge0
React

DeepScreen by Bazil Suhail

DeepScreen is an end-to-end AI hiring platform that runs from resume ingestion to offer decision in a single integrated system. It replaces spreadsheets and manual handoffs with instrumented stages, logged decisions, and consistent evaluation. The platform combines a Rust backend, Tauri desktop proctoring, Qdrant vector search, and offline AI assessment to screen, shortlist, assess, and report on candidates at scale.

Built with Rust, Tauri, Qdrant, PostgreSQL, Groq, ONNX, LiveKit, Judge0, React by Bazil Suhail, a Computer Scientist, AI Engineer, and Cross-Platform Developer based in Lahore, Pakistan.

Key Features