Bibesh Pyakurel

Software · Data · Applied AI

I build systems that connect software, data, and applied AI.

Computer science graduate from the University of Wisconsin–Green Bay. Five industry internships across backend engineering, cloud data pipelines, and AI integration, plus two arXiv preprints in computer vision and multimodal model evaluation. Currently leading applied research at UW–Green Bay.

Selected projects Read the research

$whoami→backend engineer

Open to software, backend, data, and applied-AI roles.

Portrait of Bibesh Pyakurel
Based in the United States
arXiv preprints
2
industry internships
5
best crack-detection F1
87.04%
SLAPBench fingerprint pairs
7,832

01 / Experience

Where I have worked

Five internships and a research lead role, across backend engineering, data platforms, and applied AI.

  1. Aug 2026 – PresentAshwaubenon, WI · On-site

    AI Systems Engineer · AnSer

    Building a backend engine that resolves client support requests end to end, with every automated change held for human review before it reaches production.

    • Architecting an end-to-end backend processing engine that resolves client support requests by making the configuration and code changes each one calls for.
    • Integrating multiple foundation models to parse natural-language feedback and translate stated requirements into executable logic changes.
    • Engineering secure data pipelines that apply automated updates into an isolated staging environment rather than running them against production.
    • Designing a human-in-the-loop deployment framework that holds every automated change in a staging interface for manual QA before it goes live, and provisioning the virtual machines and database tuning underneath it.
    • Backend Architecture
    • Foundation Model Integration
    • AI Agents
    • Human-in-the-Loop Systems
    • Data Pipelines (ETL)
    • Database Management
    • Virtual Machines
  2. Feb 2026 – PresentUnited States · On-site

    Research Lead · University of Wisconsin–Green Bay

    Leading applied AI research with Prof. M. G. Sarwar Murshed across pavement distress assessment and multimodal benchmark evaluation.

    • First-authored SLAPBench, the first benchmark of multimodal LLMs for four-finger SLAP fingerprint verification: 7,832 NIST SD302b pairs, five models, three prompting strategies, and a demographic fairness probe.
    • Led the pavement crack detection study to an arXiv preprint. Mask R-CNN with a ResNet-101 FPN backbone reached 84.23% precision, 90.04% recall, and 87.04% F1 on UWGB-StreetCrack.
    • Computer Vision
    • Mask R-CNN
    • Detectron2
    • Multimodal LLMs
    • Model Evaluation
  3. May 2025 – Aug 2025United States · Hybrid

    Software Developer Intern · Schneider

    Built Redis observability tooling and supported an Azure migration for offer-management applications.

    • Engineered a Spring Boot client library that collects Redis key metrics and publishes them to Dynatrace for real-time telemetry across on-premise and Azure-hosted Redis.
    • Designed integration logic that supports multiple Redis configurations dynamically, cutting deployment time by about 40% during the Azure migration.
    • Maintained the SNIConfig React application, fixing UI bugs and improving reliability in production workflows.
    • Java
    • Spring Boot
    • Redis
    • Dynatrace
    • React
    • Microsoft Azure
  4. Feb 2025 – Apr 2025United States · Remote

    Software Developer Intern · Brown County

    Built ERP backend modules on Odoo for county operations.

    • Implemented and integrated five Odoo modules in Python with PostgreSQL spanning HR, procurement, inventory, finance, and operations.
    • Refactored queries and data-access patterns to bring inter-module API response time from 125 ms to 100 ms.
    • Python
    • Odoo
    • PostgreSQL
    • ERP
  5. Sep 2024 – Dec 2024United States · Remote

    Research Assistant, Pavement Crack Detection · WiSys

    Contributed to a vision-based pavement crack detection system and its publication.

    • Collected and annotated field imagery in Label Studio to build the UWGB-StreetCrack training set.
    • Trained and compared Mask R-CNN backbone variants in Detectron2 and supported the LaTeX write-up.
    • Deep Learning
    • Label Studio
    • Detectron2
    • LaTeX
  6. May 2024 – Aug 2024United States · Hybrid

    Data Engineer Intern · Faith Technologies Incorporated

    Built Azure data pipelines, analytics dashboards, predictive models, and internal AI agents.

    • Built and optimized ETL/ELT pipelines in Azure Data Factory across five source systems, processing millions of records and cutting the reporting cycle from 3 hours to 30 minutes.
    • Delivered Power BI dashboards used by five business departments for day-to-day decisions.
    • Built an internal AI assistant with GPT-4 and Microsoft Copilot Studio that reduced employee lookup time by about 55%.
    • Azure Data Factory
    • SQL
    • Power BI
    • Machine Learning
    • Copilot Studio
  7. Feb 2024 – May 2024United States · On-site

    Computer Science Intern, Security · University of Wisconsin–Green Bay

    Ran offensive-security labs on Linux and documented them for future students.

    • Completed labs on buffer overflow, return-to-libc, ROP, format-string, and race-condition vulnerabilities on Ubuntu and Kali Linux.
    • Wrote lab manuals covering exploit setup, execution path, and mitigation strategies.
    • Linux
    • Vulnerability Analysis
    • Technical Writing
  8. Jan 2024 – May 2024United States · On-site

    Teaching Assistant · University of Wisconsin–Green Bay

    Mentored students in Python, debugging, and data analysis.

    • Led weekly lab sessions and office hours for 30+ students; the lab pass rate rose from 20% to 45%.
    • Python
    • Teaching
    • Mentoring
Also:
  • Founder & President, Nepalese Students' Association at UW–Green Bay (2024–2025)
  • Resident Assistant, UW–Green Bay Housing (2023–2025)
  • Student PEEP, The Weidner Center: ELT of ticket-buyer data into Salesforce and website updates (2023–2025)

02 / Projects

Selected projects

Products with real users and research that ships as code. Each one is on GitHub.

Full-stack productFeb 2026 – Mar 2026

ClearERP

Multi-tenant ERP for inventory and procurement workflows across six industry verticals, with per-company data isolation enforced at the database layer.

6industry verticals seeded

  • Tenant-scoped EF Core global query filters isolate every company's data; JWT auth with Admin, Inventory Manager, and Warehouse Staff roles.
  • Full purchase-order lifecycle from draft through approval, partial receipt, completion, and cancellation, with a complete audit trail.
  • Clean Architecture backend, Testcontainers integration tests against real PostgreSQL, Docker multi-stage builds, and GitHub Actions CI/CD.
  • .NET 9
  • ASP.NET Core
  • EF Core
  • PostgreSQL
  • React 19
  • TypeScript
  • Docker
CodeLive app
Full-stack productDec 2025 – Jan 2026

Trainlytics

Multi-tenant fitness analytics platform with PostgreSQL row-level security for per-user isolation and an AI insights layer on top of workout, bodyweight, and calorie data.

RLSdatabase-enforced isolation

  • Row Level Security policies guarantee zero cross-tenant reads; Supabase Auth with OTP verification and session preflight checks.
  • Dashboard analytics in Recharts for strength, muscle-group balance, and energy trends; AI insights endpoint with validation, rate limiting, and timeouts.
  • CI runs lint, typecheck, unit tests, and database migration plan validation on every push.
  • Next.js
  • TypeScript
  • Supabase
  • PostgreSQL
  • OpenAI API
  • Tailwind CSS
  • GitHub Actions
CodeLive app
Research benchmark2026

SLAPBench

The first benchmark for evaluating multimodal large language models on four-finger SLAP fingerprint verification, built on NIST SD302b and released with code and pair manifests.

7,832fingerprint pairs evaluated

  • 7,832 exhaustive pairs (176 mated, 7,656 non-mated) across InternVL3-8B, Qwen2.5-VL-7B, Qwen3-VL-8B, Gemma-3-12B, and Claude Opus 4.8.
  • Shows prompt-driven verification collapse under binary prompting and recovers discrimination with similarity-scoring prompts.
  • Includes matched-resolution and demographic fairness probes over gender, race, and age.
  • Python
  • PyTorch
  • Multimodal LLMs
  • NIST SD302b
  • Evaluation
CodePaper
Chrome extension + backendJan 2026 – Feb 2026

JobInsight

Chrome extension that analyzes LinkedIn job postings in real time and surfaces sponsorship, citizenship, experience level, and keywords in a draggable overlay.

<$0.0005cost per job analysis

  • Chrome MV3 architecture with content script, service worker, and popup UI; keywords highlighted directly in the job description.
  • Node.js/Express proxy keeps the OpenAI key server-side behind a Google OAuth gate; seven-day cache makes repeat views instant.
  • GPT-4o-mini keeps each analysis under $0.0005.
  • Chrome MV3
  • JavaScript
  • Node.js
  • Express
  • OpenAI API
  • Google OAuth
Code

More projects

  • Research

    Pixel-Level Pavement Distress Assessment

    Mask R-CNN instance segmentation on the field-collected UWGB-StreetCrack dataset; the ResNet-101 FPN backbone reached 87.04% F1.

    2024 – 2026Python · PyTorch · Detectron2
  • Machine learning

    Music Emotion Predictor

    Classifies 278K+ tracks into Calm, Sad, Energetic, and Happy with Decision Tree, KNN, and K-Means; evaluated with accuracy and Adjusted Rand Index, visualized with PCA.

    Jan 2025 – May 2025Python · scikit-learn · Pandas
  • NLP automation

    Resume–Job Description Matcher

    Parses PDF resumes, scores them against a job description with keyword matching and cosine similarity, and batch-ranks candidates.

    Sep 2024 – Dec 2024Python · scikit-learn · NLTK
  • Web tool

    Script Viewer

    Parses call-center script XML into JSON and renders it for review while preserving every element, attribute, namespace, and sibling order.

    2026TypeScript · Vite
  • Security

    Cybersecurity Labs

    Hands-on buffer overflow, return-to-libc, ROP, and format-string exploitation on Linux, written up as reusable lab manuals.

    Spring 2024Linux · Kali · GDB
  • Desktop app

    Typing Speed Game

    Monkeytype-inspired JavaFX trainer with real-time WPM, accuracy, and error tracking in an MVC structure.

    2024Java · JavaFX
  • Desktop game

    Blackjack

    Collaborative Java desktop Blackjack with a card engine, betting logic, and Scene Builder GUI.

    Fall 2023Java · Scene Builder
  • Desktop app

    Student Budget Calculator

    Java budgeting tool with category grouping, remaining-balance tracking, and input validation.

    Fall 2023Java · Scene Builder
  • Mobile

    Smart Student App

    Android utility app for students with native features such as one-tap emergency calling.

    Fall 2023Android Studio · Java
  • Database design

    Online Service Database

    ER modeling, normalized schema with constraints, and a SQL query set for an online service system.

    Fall 2023SQL · ER modeling

03 / Research

Papers

Applied computer vision and multimodal model evaluation, with an emphasis on rigorous benchmarks and reproducible conclusions.

202601

Jul 17, 2026 · arXiv preprint · cs.CV, cs.AI · first author

SLAPBench: Benchmarking Multimodal Large Language Models for Four-Finger SLAP Fingerprint Verification

Bibesh Pyakurel · M. G. Sarwar Murshed

Introduces the first benchmark for MLLM-based four-finger SLAP fingerprint verification on NIST SD302b: 7,832 pairs, five models, and zero-shot, task-description, and similarity-scoring prompts. Shows prompt-driven verification collapse and includes a demographic fairness probe.

arXivPDFDOICode & data
202602

May 25, 2026 · revised Jul 13, 2026 · arXiv preprint · cs.CV

Pixel-Level Pavement Distress Assessment Using Instance Segmentation

Logan Dewick · Bibesh Pyakurel · Kong Pheng Yang · Nazim Choudhury · M. G. Sarwar Murshed

Evaluates Mask R-CNN instance segmentation on the field-collected UWGB-StreetCrack dataset across longitudinal, transverse, and alligator cracks plus potholes. The ResNet-101 FPN backbone reached 84.23% precision, 90.04% recall, and 87.04% F1, with predicted crack-area fraction within 0.006 points of ground truth.

arXivPDFDOI
Explore the research portfolio Interests: Computer vision for infrastructure · Multimodal model evaluation · Trustworthy biometric AI

04 / Skills

What I work with

Backend & APIs

  • Java
  • Spring Boot
  • C#
  • .NET 9
  • ASP.NET Core
  • Python
  • Node.js
  • Express
  • REST APIs
  • Redis
  • PostgreSQL
  • SQL Server
  • Supabase
  • JWT / OAuth 2.0
  • Row Level Security
  • RBAC

Data Engineering

  • Azure Data Factory
  • ETL / ELT
  • PySpark
  • Pandas
  • Data Modeling
  • Data Warehousing
  • Power BI
  • Snowflake
  • BigQuery

AI & Machine Learning

  • PyTorch
  • TensorFlow
  • scikit-learn
  • Detectron2
  • Mask R-CNN
  • Multimodal LLMs
  • Model Evaluation
  • OpenAI API
  • Label Studio

Cloud, DevOps & Web

  • Microsoft Azure
  • AWS
  • Docker
  • Kubernetes
  • GitHub Actions
  • Dynatrace
  • React
  • Next.js
  • TypeScript
  • Tailwind CSS

05 / About

Trained as an engineer, poet at heart

I build systems that connect software, data, and applied AI: production-oriented APIs and data pipelines, multi-tenant web applications, and computer-vision research that ships as reproducible code.

At Schneider I built a Spring Boot library that publishes Redis metrics to Dynatrace across on-premise and Azure environments. At Faith Technologies I moved millions of records through Azure Data Factory pipelines into Power BI dashboards. On my own time I ship multi-tenant products with database-enforced isolation.

I care about why a system exists as much as how it works. In 2020 I published यथार्थ (Reality), a Nepali book on love, society, and inequality. It taught me that empathy and nuance matter as much as functionality, and that the best solutions connect with people.

Bibesh Pyakurel in graduation regalia outside the Weidner Center at UW–Green Bay
Graduation, December 2025 · UW–Green Bay
Sep 2022 – Dec 2025

B.S. Computer Science · University of Wisconsin–Green Bay

  • Dean's List with High or Highest Honors, six semesters
  • Senior research: instance segmentation for pavement distress assessment, WiSys-funded
  • Coursework: AI, Data Structures & Algorithms, Database Design, Cloud Computing, Advanced Software Design, Theory of Programming Languages

Author

Published यथार्थ (Reality) in 2020, a Nepali book exploring love, society, and inequality.

Community

Founded and led the Nepalese Students' Association at UW–Green Bay. Three years as a Resident Assistant.

Teaching

Teaching Assistant for Python labs, mentoring 30+ students and helping raise the lab pass rate from 20% to 45%.

06 / Contact

Let's build something that matters.

Open to software, backend, data, and applied-AI roles. Email is the fastest way to reach me; LinkedIn works too.

· Semantic Scholar · United States