AI Engineer · Co-founder, Khyontek AI

Dr. Pritam Deka

Now
Co-founding Khyontek AI (Guwahati) ARC research: agentic process extraction with VLMs 2 new papers at ACM SAC 2026 & IEEE Access

About

I am an AI Engineer and AI Research Fellow specialising in practical AI systems across LLMs, RAG, agentic AI, multimodal AI and document intelligence. My research focuses on fact verification of health information, information extraction using large language and vision-language models, and AI for document and process understanding. I thrive on turning research ideas into working systems, demos and reusable pipelines, and on contributing to the open-source community. I also co-founded Khyontek AI, building SME-focused AI products.

Research Interests

Natural Language Processing Large Language Models Vision-Language Models Information Extraction Health Information Fact-Checking Named Entity Recognition Transformer Models Prompt Engineering Multimodal AI Health Informatics Code-Mixed NLP AI for Business Process Mining Language Endangerment

Research Topics Network

Interactive map showing connections between my research areas and collaborators. Hover over nodes to explore, drag to rearrange.

Core Research
Applications
Methods
Domains
Collaborators

Latest News & Updates

January 2026

Co-Founded Khyontek AI

Co-founded Khyontek AI (Guwahati, Assam) as Co-Founder & AI Product Lead, building SME-focused AI products and leading product strategy, fundraising preparation and an AI internship programme.

March 2026

Paper Accepted at ACM SAC 2026

Our work on "Structured Extraction from Business Process Diagrams Using Vision-Language Models" has been accepted for presentation at ACM SAC 2026.

January 2025

Vision and Language Symposium Speaker

Main speaker at the Vision and Language Symposium 2025 held at Queen's University Belfast.

May 2024

Started as Research Fellow at QUB

Joined Queen's University Belfast as a Research Fellow in AI, focusing on corporate document understanding with LLMs and VLMs.

December 2024

PhD Completed

Successfully defended PhD thesis on "Evidence-Based Approach to Verification of Online Health-Related Content" at Queen's University Belfast.

Academic Journey

Skills

Python
NLP
LLM/VLM
Data Pipelines
Information Extraction
Model Evaluation
Research

Projects & Live Demos

Flowchart2Mermaid

Vision-Language Model powered app converting uploaded flowchart images into editable Mermaid code, with diagram preview, export and AI-assisted refinement.

VLMs Vercel Process Mining

BelfastBuild AI

RAG-based planning-compliance pre-screener for Northern Ireland planning proposals. Cross-references policy sources, validates postcodes, flags conflicts and cites sources.

RAG Vercel Retrieval

Biomedical Fact-Checker

PhD-derived demo verifying health-related claims against biomedical evidence using retrieval, evidence extraction, claim classification and explanation generation.

BERT Gradio Health AI
Internal

BPMN Fullstack App

Fullstack image-to-BPMN XML prototype with a FastAPI backend, OpenAI GPT-4o and a frontend editor for image upload and live BPMN rendering. Built for PwC ARC.

FastAPI GPT-4o BPMN

Cyber-Attack Attribution Dataset

Publicly released labelled dataset for named entity recognition and predictive cyber-attack attribution, with preprocessing and evaluation workflows.

NER Cybersecurity Dataset

Open-Source AI Assets

50+ fine-tuned models and 20+ datasets on Hugging Face across document intelligence, retrieval, classification, biomedical NLP and multimodal reasoning.

Hugging Face Open Source

Experience

Co-Founder & AI Product Lead

Jan 2026–Ongoing | Khyontek AI, Guwahati, Assam
  • Lead product strategy and hands-on AI engineering for SME-focused AI products, translating business problems into feasible solution concepts, prototypes and implementation plans
  • Built a structured AI internship programme and mentor students on practical AI workflows including problem scoping, dataset preparation, model experimentation and reproducible delivery
  • Lead fundraising preparation, product positioning and go-to-market planning, balancing technical ambition with customer value, implementation cost and commercial viability

Research Fellow (AI)

May 2024–Ongoing | Queen's University Belfast
  • Focusing on AI for corporate document understanding with multimodal data processing
  • Investigating GPT-based models and open-source LLMs/VLMs for corporate data analysis
  • Integrating NLP and prompt engineering to streamline information extraction
  • Evaluating prompts and models on curated process diagram datasets
  • Developed agentic AI workflows converting flowchart and process diagram images into executable diagram code, supporting process standardisation and documentation.
  • Delivered high-accuracy multimodal AI pipelines by fine-tuning open-source VLMs (Qwen2.5-VL, Qwen3-VL, Gemma 3) and applying advanced prompt and context engineering with frontier LLM APIs (GPT-4.1, GPT-5.2, Gemini 2.5 Flash), outperforming OCR-centric approaches by a wide margin.

Subject Teacher

May 2025–August 2025 | INTO Queen's University Belfast
  • Delivered lectures in Object-Oriented Programming using C++
  • Designed assignments, assessments, and practical coding labs
  • Implemented student engagement techniques and formative assessment tools

Senior Research Assistant (AI-Security)

Jul 2023–Jan 2024 | University of Southampton
  • Led a team to create a NER dataset for cyber-attack attribution
  • Developed methods and metrics for dataset quality assessment
  • Ensured dataset suitability for training NER models in cybersecurity

AI Intern

Mar 2022–Sep 2022 | Momentone
  • Built a mental health chatbot using Rasa and HuggingFace transformers
  • Developed NLP solutions for mental health support

Lecturer

Aug 2018–Sep 2019 | Royal Global University, Assam
  • Taught Python, C++ OOP, Data Structures, and Algorithms
  • Supervised undergraduate projects and contributed to curriculum development

Get In Touch

Dr. Pritam Deka