AI should speak
every language.
I'm Shah Bakhsh. I build intelligent systems across language models, machine learning, and agentic AI. One mission drives much of my work: helping bring Balochi into the future of technology.
A personal workspace for experiments, research, and things worth building.
A BELIEF THAT BECAME A BODY OF WORK
Some languages are missing from the future.
I want to help change that.
A language should not need millions of speakers to deserve a place in technology. It deserves tools, data, models, and the chance to be understood.
THE MISSION / BALOCHI AI01 / The person behind the systems
Curiosity,
made useful.
I'm an AI engineer focused on language technology, applied machine learning, and intelligent systems that solve meaningful problems.
A major part of my work is centered on Balochi, a low-resource language that deserves better digital tools. I work across data preparation, tokenization, model training, evaluation, and usable NLP applications. The goal is not simply to train models. It is to help build the foundations that let a language participate in the digital world.
I also build with LLMs, retrieval systems, and AI agents, looking for ways to turn research into dependable tools. I care about clear evidence, open-source work, and engineering that makes a difference beyond a demo.
02 / Technical range
My tech superpowers.
A practical toolkit for taking ideas from datasets and experiments to usable AI-powered products.
Applied AI & LLMs
Model adaptation, prompt systems, evaluation, and practical LLM applications.
Language technology
Low-resource NLP, tokenization, linguistic datasets, and multilingual systems.
Agentic systems
Tool use, multi-agent workflows, retrieval, and task orchestration.
Computer vision & OCR
Image understanding, document processing, and OCR experimentation.
03 / A mission worth exploring
The Balochi AI Lab.
Building blocks for language technology, from foundational text processing to machine learning experiments.
INTERACTIVE RESEARCH EXPLORER ↓
RESEARCH MISSION CONTROL
BAL-NLP / EXPLORER INTERFACE
SELECTED RESEARCH NODE
Choose a research node
Select any node on the left to inspect a part of the Balochi AI research workspace. Each selection reveals a short project overview.
TRY IT: Select a research node to reveal its mission briefing.
04 / Selected work
Built to explore.
A selection of research projects and experiments. Open the repositories for source code, implementation details, and current project status.
BalBERT
Domain-adaptive language representation work focused on strengthening NLP foundations for Balochi.
BalTokenizer
A Unigram tokenizer built for Balochi text, with documented character coverage and unknown-token evaluation.
BalPOS
A transformer-based Balochi part-of-speech tagging project built around Universal Dependencies.
BalNER
Balochi named-entity recognition, with documented entity-level evaluation and dataset limitations.
BalNLP
An Arabic-script Balochi NLP workspace bringing together a Python library, FastAPI service, and right-to-left interface.
CrisisPilot AI
An AI-driven concept for disaster intelligence and emergency coordination workflows.
PathPilot AI
A career navigation platform exploring personalized roadmaps, resume analysis, and AI mentorship.
ReelForge
An experimental video clipping workflow combining speech recognition, computer vision, and local language models.
05 / The next collaboration
Have a meaningful problem? Let's build.
I'm interested in research, open-source collaboration, and building AI tools with real-world value. Reach out and let's start a conversation.
Contact me