About Me
Hello! I’m Bibek from Nepal. I enjoy the process of creating complex applications and I love experimenting with new open-source libraries and exploring innovative B2C tech products. My expertise includes managing large-scale project backends, creating user-friendly Python libraries and APIs, handling network gateways for SaaS platforms, containerizing and packaging applications, developing robust document parsers and LLM apps like RAGs, and crafting quick demos with intuitive frontends and UX interactions.
I’m a firm believer in my mantra, “If they can Do it, I can Do it”.
If you are bored to read this entire page, feel free to chat with my chatbot widget at bottom right.
Professional Skills
Python
MLflow
Git
PyTorch
JavaScript
DVC
Docker
HuggingFace
Nginx
Ollama
TorchServe
GitHub Actions
Linux
FastAPI
LlamaIndex
Gradio
Kong
GoogleCloud
Work Experience
AI Researcher at Artificial Language Intelligence (ALI Co. Ltd), Korea
Aug, 2023 - Present
DocumentAI Research
- Develop neural-network-based Layout Detector (306M), labels: [text, title, list, figure, table]
- Smart augmentation techniques with controlled noise injection specific for Document tasks that make the model language independent
- A ranker for natural reading order handling complex multi-columned documents
- Two versions: fast (TensorRT-10FPS in T4GPU) vs slow (Openvino-1FPS in CPU) for BATCH_SIZE=1
- Develop fast-header detection model for selectable PDFs that uses font-attributes (name, color, flag), drawings (lines & rects) and header regex patterns.
- Maintainer of ALIParse (in-house python library) that provides easy access to layout analysis modules, independent of document types and extends to provide granular APIs and RAG service
- Serve Layout Detector models using Torch-serve, local LLMs using Ollama/Vllm and setup automated workflows using GitHub Actions
- Set up reverse proxies for all demo applications and manage the Kong gateway
- Highly familiar with Gradio and setting up quick demo applications on HuggingFace space
- Implement data versioning using DVC; setup local Model registry that supports model versioning and experiment tracker; make Kubeflow training pipelines
- Prepare executables for Python and Node.js applications to run on Windows and macOS. Additionally, obfuscate and make installer for these executables.
- Hands-on experiments to optimize and run heavy models on CPU/GPU like converting models to OpenVino, ONNX, and TensorRT.
- Secured 5th position in AI Grand Challenge 2023 (agc2023). My role was to make a Docker image that runs 4 models (1 polyglot and 3 Llama) efficiently within a 40GB image size constraint for offline inference on V100 GPUs.
Layout Detector
Document Parser
Document RAG
Form Extraction
Student Assistant at International Scholar and Student Services (ISSS), KAIST
Mar, 2023 - Present
Assisting international students and visitors who visit the International Student and Scholar Services (ISSS) office by providing them with accurate information and guidance. My role is to ensure that visitors receive the appropriate support and are directed to the appropriate staff members as needed. I also volunteer in preparation and organization of different international community's events like KAISTONE.
Researcher Intern at Robust Intelligence & Robotics Lab, KAIST
Jul, 2022 - Dec, 2022
Worked on developing a combined framework for realtime 6D Pose Estimation and Scene Graph Generation (SGG) in Robotic Manipulation Task. I used Cosypose framework for 6D pose and Neural Motif framework for SGG and performed experiments on custom cup datasets.
- For a single moving camera setup, estimated 6D pose and generated dynamic scene graphs.
- For 2 fixed camera setup, developed a probabilistic multi-view object-pose estimation framework that (i) associates multiple estimation results with scene graphs, (ii) combines pose distributions from single-view based estimators, and (iii) constructs a unified scene graph by predicting a unified pose distribution per object using MC dropout.
- Demo:
Backend Developer Intern at Bisonai, Seoul
Jan, 2022 - Feb, 2022
- Built data pipelines (DAGs) in Airflow to collect DeFi data from various sources like Uniswap, Coingecko and stored them in MongoDB Atlas.
- Implemented backend API server using FastAPI and Strawberry (Python GraphQL Library) to provide DeFi data from MongoDB for users interested in knowing DeFi data from multiple Decentralized Exchanges.
- Challenged by requirement to gain an in-depth understanding of raw DeFi data and their importance to Bisonai’s ML based System Trading project.
Project Intern at Interactive Computing Lab,
KAIST
Jul, 2021 - Aug, 2021
- Made a simple tutorial for data pre-processing and visualizing plots with Vegalite and D3.js in ObservableHQ notebook for beginners to learn Data Visualization techniques using Javascript track.
- Reference: observablehq.com/@ic-dataviz?tab=notebooks
Coordinator and Tutor at A Levels Ex-students Tutoring Center (ETC)
Jul, 2018 - June, 2019
- Taught home-tuitions on hourly basis for A-Level students on Chemistry, Physics, Mathematics and Biology. I connected to students by publishing advertisements on hamrobazaar.com and asking for referrals of my teachers.
Teaching Assistant at British Model College
(BMC)
Jul, 2017 - Oct, 2017 and Feb, 2018 - May, 2018
- Conduct tutorial lectures and helping sessions- help/guide students if they have problem understanding any topics of the course (Chemistry, Physics, Mathematics and Biology)
- Counseled students giving examination tips and college application suggestions.
Others:
- Part-time A Levels Tutor at BNS Education Hub, New Road, Kathmandu [Jan, 2018 - Jun, 2019]
- Class Representative (CR) in British Model College for 2 years
- Prefect of Swarnim School during class 10
- Executive Member of Global Peace Young Leaders (GPYL) [Jun, 2018 - Jun 2019]
- General Member of Society of Nepalese Students in Korea (SONSIK) [Sep, 2021 - Sep, 2022]
Education
Bachelor of Electrical Engineering and Computer Science from KAIST, Daejeon, South Korea
(2019 - 2023)
GCE A Levels from British Model College, Kathmandu,
Nepal
(2015 - 2017)
Grade 8-10 from Swarnim School, Kathmandu, Nepal
(2013 - 2015)
Grade Nursery-8 from Khimti Project School, Dolakha,
Nepal
(2002 - 2013)
Achievements
- Awarded KAIplus+ scholarship for Spring Semester 2023
- Won the Social Innovation Award in SDG Hack 11, TSINGHUA GLOBAL SUMMER SCHOOL 2021 with topic 'AI to Solve Education Accessibility Problems in Rural Areas'
- Common Purpose Alumni and awarded the badge "Inclusive Leadership 2021"
- Awarded KAIST International Student scholarship with stipends for 4 years [2019-2023]
- Gained the highest mark in Nepal for AS Level Chemistry in Nov, 2016 examination series
- Secured 2nd position in Quiz in Budhanilkantha Inter-School Olympiad, 2016
- Secured 3rd position in Cassini Scientist For a Day Contest, 2017
- Best student of the batch 2015/2017 in British Model College
KAIST Activities
- Student Assistant in International Scholar and Student Services (ISSS) team at KAIST from Spring 2023
- Candidate for 2019-Winter Alma Mater visiting program and conducted KAIST info session at my high-school: LINK
- Video won the ‘Excellence Award’ in 2020 Happy KAIST UCC Contest to aware about mental health: LINK
- Paper (Idea) won the ‘INSIGHTFUL’ award in the online Public Relations strategy contest, 2020 for KAIST admissions to engage in international communities
- Second winner on the writing contest, ‘What is like studying at KAIST?’ organized by the KAIST Admission Office in December, 2019
- International Community Representative of Nepal for Spring 2020, Spring 2021, Fall 2021, Spring 2022 and Fall 2022
- Online Mentor for new international students of Fall 2020, Spring 2021, Fall 2021 and Spring 2022
- Graduated with 142.0 credits and list of courses taken each semester: course-records