Speech AI · Library
auralis-vfs
Pip-installable Python library for vocal fatigue scoring from raw speech. Built on ECAPA-TDNN-VHE — returns a normalized fatigue score from any audio input.
PyPIECAPA-TDNNSpeech
Applied AI Lab · Multan, Punjab, Pakistan
Research. Engineering.
Education. Built to ship.
INFERENCE Lab conducts reproducible AI research, builds production systems, and trains engineers who can deploy — not just describe. Engineering discipline over hype, evidence over branding.
- Largest
- ROMAN URDU LANGUAGE RESOURCE
- SOTA
- MODELS FOR LOW-RESOURCE LANGUAGE & HEALTH RESEARCH
- GLOBAL
- RESEARCH COLLABORATIONS & PUBLICATIONS
- 4+
- OPEN-SOURCE AI LIBRARIES
What we are
One organization, three reinforcing tracks.
INFERENCE Lab is an applied AI research and engineering organization founded by Muhammad Khubaib Ahmad. It operates as original research, AI/software engineering services, and structured engineering education — under one identity, each track reinforcing the others' credibility.
Our aim: close the gap between people who know AI concepts and engineers who can build, deploy and maintain AI systems.
Engineering discipline over hype
Real, deployed output — not demos that break the moment they leave a notebook.
Evidence over branding
Reproducible pipelines, proper evaluation, and a permanent DOI on every research release.
Output over certificates
You leave with systems on GitHub and models on HuggingFace, not a PDF.
What we do
Three tracks, one engineering standard.
Research informs the curriculum. The curriculum trains the engineers. The engineers ship the services. Each track makes the others credible.
01
Research
Original research in low-resource NLP and speech intelligence — conducted independently and through international academic collaboration.
- King Saud University · EPU Kuwait · Doane University (USA) · Hanyang University (Korea)
- Every release ships reproducible pipelines + a permanent DOI
- Deployable inference code, not leaderboard numbers
02
Engineering Services
AI/ML engineering, LLM and RAG systems, computer vision, speech AI, and full-stack delivery for organizations.
- Production REST APIs, containerized & monitored
- Same engineering discipline taught in the lab
- Owned end-to-end across the full ML lifecycle
03
Engineering Education
A structured, deployment-focused curriculum that takes developers from basic Python syntax to deploying production AI systems.
- 12.5 months · 6 phases · live coding every week
- Engineering Fellowship for project-based contribution
- Built, deployed, and on GitHub — not certificates
Research output
Low-resource NLP & speech intelligence.
Independent and international research — every release ships reproducible pipelines, evaluation documentation and a permanent DOI.
- Under Review2026
Modeling Vocal Fatigue as Embedding-Space Deviation Using Contrastively Trained ECAPA-TDNNs
ECAPA-TDNN-VHE designed from scratch with supervised contrastive loss — 2.5× accuracy over baseline (78% vs 36%), F1 scores 0.85 / 0.78 / 0.70 across three fatigue classes.
Springer · EURASIP J. on Signal ProcessingDOI - Under Review2026
Continuous Vocal Load Monitoring in Professional Voice Users
Development and occupational validation of an automated vocal load assessment tool for professional voice users — clinical-grade speech analysis in production.
Journal of Voice · King Saud University & EPU Kuwait - Under Review2026
RUEmoCorp: A Large-Scale Roman Urdu Emotion Corpus & Benchmark Suite
First large-scale Roman Urdu emotion corpus — 134K labeled samples with Fleiss κ = 0.658 (substantial agreement), multi-institute annotation, fully open-source on HuggingFace and Harvard Dataverse.
Language Resources and Evaluation (Springer)DOI - Published Preprint2026
RUDaSA: Roman Urdu Dataset for Sentiment Analysis — A Large-Scale, Curated Corpus with Privacy-Preserving Embeddings and Competitive Benchmarking of Transformer Models
Large-scale Roman Urdu sentiment corpus built via privacy-preserving embedding pipelines. Benchmarks state-of-the-art Transformer models — addressing a critical gap in low-resource South Asian NLP.
Research Square · PreprintDOI - Published Preprint2025
Data-Centric Roman Urdu NLP: Dataset Curation & Model Benchmarking
Largest high-quality Roman Urdu sentiment dataset via privacy-preserving embedding pipelines — SOTA 0.84 accuracy, 0.83 Macro-F1.
Zenodo · PreprintDOI - Published Preprint2025
Forecast-Based Decision Support System for Mango Malformation
Time-series forecasting and smart-agriculture DSS — demonstrated 50–60% yield improvement through data-driven intervention.
Zenodo · PreprintDOI - In Progress2026
Ergonomic Interventions and Cognitive Workload in Healthcare Settings: A Qualitative Case Study Using Cognitive Systems Engineering
Multi-institutional international study applying Cognitive Systems Engineering to healthcare ergonomics — systematic analysis of workload, safety, and intervention efficacy.
Hanyang University (Korea) · King Saud University (Saudi Arabia) · Doane University (USA)
Software & systems released
Open source, deployed, and used by researchers.
Libraries, APIs and platforms spanning speech AI, LLM observability, RAG, agents and MLOps infrastructure.
Speech AI · Real-Time
voiceMonitor
Real-time vocal health monitoring tool — streams microphone input, runs auralis-vfs, and surfaces fatigue level and vocal load over a session.
Real-TimeMonitoringCLI
Speech AI · Verification
VocalID
Open-source speaker verification framework using embedding cosine similarity on ECAPA-TDNN representations. Lightweight, no external APIs required.
Speaker VerificationEmbeddingsOpen Source
Open Source · Library
faker-pk
Pakistani synthetic data library for realistic test fixtures — Urdu and Roman Urdu names, CNIC numbers, phone formats, addresses, and locale-aware records.
PyPISynthetic DataPakistan
LLM Observability
QueryVault
End-to-end LLM platform with real-time hallucination detection, structured logging, and a developer dashboard — hallucination rate, latency, P50/P95.
StreamlitFew-ShotObservability
Multi-Agent System
DataForge
Multi-agent system for automated EDA, cleaning, visualization, and report generation — autonomous data analysis pipelines built on CrewAI.
CrewAIEDAAgents
Image Encryption
SecureCipher v2.0
Image encryption scheme scoring 100/100 against 14 benchmark algorithms across standard cryptographic security tests.
CryptographyBenchmarked
Engineering education
Full curriculumFrom Python syntax to deployed AI systems.
A structured, deployment-focused curriculum across 6 phases and 12.5 months. Every 1.5-hour session is live coding — Saturday introduces the concept, Sunday goes deeper into edge cases, and assignments ship to GitHub by Friday.
- Phase 02 months
Engineering Foundations
From a working terminal to production database architecture. Writing Python that does not break.
8 weeks
- Phase 12 months
Data Engineering & Visualization
NumPy through publication-quality visualization, grounded in real statistical thinking.
8 weeks
- Phase 22.5 months
Machine Learning Engineering
Classical ML the engineering way: pipelines, leakage-free evaluation, and experiment tracking.
7 weeks
- Phase 32.5 months
Deep Learning & NLP
Neural networks from scratch to fine-tuned Transformers and speech AI pipelines.
8 weeks
- Phase 42 months
AI Systems & LLM Engineering
Production AI APIs, disciplined LLM engineering, RAG systems, and agents that actually work.
4 weeks
- Phase 51.5 months
MLOps & Deployment
Containerize, automate, ship and monitor. The difference between a notebook and a product.
3 weeks
The founder
Muhammad Khubaib Ahmad
AI Research Engineer
An AI research engineer at the intersection of rigorous research and production engineering — designing architectures from scratch, publishing reproducible research with open-source artefacts, and deploying end-to-end ML systems independently. Mentors and trains engineers in applied AI.
Speech & Language IntelligenceLow-Resource NLPLLM EngineeringContrastive Learning
RoleFounder & Director, INFERENCE Lab
CollaborationsKing Saud University · EPU Kuwait · Doane (USA) · Hanyang University (Republic of Korea)
Engineering Fellowship
Contribute to real, shipped work.
Beyond the cohort, the Fellowship is hands-on, project-based contribution to the lab's open-source surface. The first Fellowship runs across three active projects — you write code that real users depend on.
First cohort live · Python Engineering phase
- 01
faker-pk extension
Extend the Pakistani synthetic data library with new locales, providers and record types.
- 02
Lab website
Design and ship the public-facing engineering surface for the lab and its open-source work.
- 03
Dataset quality auditor
A tooling project to automatically profile, score and flag quality issues in research datasets.
Join INFERENCE Lab
Work on real AI systems, research, and open-source.
We're building AI systems, conducting research, and training engineers. Every position involves real, shipped work.
Engineering Fellowship
3-Month · Remote · Flexible
Work on real AI systems alongside the lab — agentic pipelines, LLM applications, speech AI, and production MLOps. Designed for engineers who want serious project experience, not busy-work.
Volunteer Research Contributor
Rolling Applications
Contribute to open-source tooling, dataset annotation, or benchmarking tasks on a flexible schedule. No minimum commitment — just genuine interest in doing useful work.
Industry Collaboration
Project-Based · Open
Partner with INFERENCE Lab on applied AI R&D. We work with organizations that need rigorous, reproducible engineering — not a vendor relationship, a research partnership.
Work with the lab
Train as an engineer, collaborate on research, or build a system with us.
Currently running our first online cohort and preparing the first Engineering Fellowship. Reach out about training, research collaboration, or AI engineering services.