Bishwash Khanal

Project Researcher - ANSE Project

University of Jyväskylä

Research assistant Jan–Jun 2026; promoted to project researcher in Jul 2026.

• Build and evaluate multi-agent systems for code generation under test-driven development, orchestrating LLM agents with LangGraph and gating each step on automated Pytest runs
• Design and iterate on TDD orchestration workflows for AI-native software engineering
• Embed traceability, automated verification, and human-oversight checkpoints into agentic development workflows, aligning with EU AI Act and ISO/IEC 42001 requirements
• Researched failure modes in agentic TDD pipelines and proposed practical mitigations; co-authored TDD Governance for Multi-Agent Code Generation, published at Prompt SE @ EASE 2026
• Technologies: Python, LangGraph, OpenAI API, Pytest, Docker

Artificial Intelligence Engineer

OptiML - founding team member, part-time

• Developed custom pipelines for fine-tuning and compressing open-weight LLMs (Llama, Qwen, Mistral, Gemma, Phi, DeepSeek), applying SparseGPT, Wanda, and magnitude pruning at up to 50% sparsity - showing that task-specific calibration data significantly improves downstream performance of compressed models
• Evaluated state-of-the-art LLM compression techniques and published the results as Evaluating the Impact of Compression Techniques on Task-Specific Performance of Large Language Models
• Integrated model compression into the fine-tuning workflow to support efficient deployment at scale
• Set up the training stack, experiment tracking, and evaluation harness as founding engineer
• Technologies: PyTorch, Transformers, PEFT, BitsAndBytes, Accelerate, Weights & Biases

Computer Vision Engineer

E.K. Solutions Pvt. Ltd.

• Quantized and deployed face detection, recognition, and swapping models (YOLOv8, ArcFace, Inswapper, StyleGAN2) on Jetson AGX Orin with INT8/FP16 quantization via PyTorch, ONNX Runtime, and TensorRT - reaching up to 36 FPS at 640×480, roughly 2× the FP32 CUDA baseline
• Developed an early-adoption RAG chatbot integrating GPT-4 for project management, database analysis, and bulk CV screening using LangChain, ChromaDB, and FastAPI, deployed internally across a 300-employee organization
• Generated planar-simplified 3D meshes for virtual tours from iPad Pro LiDAR captures, implementing custom algorithms for planar simplification, ICP registration, TSDF volumetric integration, and MVS texturing
• Applied RandLA-Net for semantic segmentation and built the full-stack pipeline in Python, C++, Swift, and Three.js
• Technologies: PyTorch, TensorRT, ONNX, OpenCV, Open3D, LangChain, FastAPI, Celery, NumPy

Freelance & Contract Work

GoThru Media · Scale AI

Computer vision automation for GoThru Media - automated pose estimation between uncalibrated spherical panoramas, removing a manual production step. LLM training-data evaluation for Scale AI.

Teaching Assistant

Jacobs University Bremen

Supported students in electronics and communications labs, graded assignments, and ran tutorials alongside my studies.

• Intro to Electronics: KiCAD, LTSpice, PCB simulations
• Communications Lab: C and MATLAB for DSP and communications

Student Supervisor

IRC Library, Jacobs University Bremen

Supported students and patrons at the university library and trained new employees.

M.Sc. in Artificial Intelligence

University of Jyväskylä

Grade 5/5. Awarded the JYU Rector's Scholarship for exceptional academic achievement.

• Thesis: From General Description to Expert Critique - Modular LoRA Adapters for Domain-Specific Photographic Assessment
• Deep Learning for Cognitive Computing, Natural Language Processing, Computer Vision and Image Analysis, Semantic Web, SOA and Cloud Computing

B.Sc. in Electrical & Computer Engineering

Jacobs University Bremen

Grade 1.73, minor in Intelligent Mobile Systems. Merit-Based Scholarship for academic achievement.

• Thesis: Non-Linearity in Wireless Communications and Deep Learning
• Machine Learning, Computer Vision, Artificial Intelligence, Signal Processing, Robotics