Thesis: Hybrid Multi-Agent Orchestration – Evaluating Gas Town & Gas City in Enterprise Environments

High Level Description
The software development lifecycle is shifting from traditional human-centric coding to highly automated, agentic workflows. Prominent open-source frameworks like Gas Town and its modular counterpart Gas City act as orchestrators for dozens of parallel model runtimes (such as Claude Code, Gemini, and local LLMs) to autonomously plan, implement, test, and merge enterprise code.

Deploying such massive agentic architectures within an enterprise environment introduces complex infrastructure, data privacy, and cost challenges. Relying solely on external commercial models can lead to high token costs and data leakage risks. Conversely, relying only on local compute can bottleneck performance.

This thesis will investigate the business and technical viability of adopting Gas Town/Gas City inside a hybrid enterprise infrastructure. The project will center around building an enterprise-grade Proof of Concept (PoC) running locally and in private cloud environments, evaluating how different local and frontier models can be strategically routed to balance cost, performance, and data security.

Project Description
The core objective of this project is to benchmark the feasibility, economic cost, and architectural flexibility of Gas Town/Gas City for enterprise-level software workflows across a hybrid hardware topology. The student(s) will design and deploy a contained Proof of Concept (PoC) factory that operates primarily on local developer machines and enterprise private cloud infrastructure.

A central part of the research will be experimenting with heterogeneous model routing based on compute constraints. Instead of running uniform high-cost frontier models everywhere, the student will map out and benchmark the performance of routing specific tasks to a tiered hardware and model infrastructure:

• Lightweight Tasks: Using small CPU-optimized models and small GPU-accelerated local models for basic parsing, formatting, or simple worker tasks (e.g., Polecat/Worker roles).
  
• Medium/Heavy Local Tasks: Utilizing large local open-weights models deployed on high-end local hardware (e.g., 128GB GB10 Grace Blackwell architecture) for complex local reasoning, code synthesis, and contextual search.
  
• Complex Orchestration: Using commercial or private-cloud hosted frontier models for high-level management, orchestration (e.g., Mayor/Coordinator roles), and structural code validation.
  

Who are we looking for?

• Bachelor or Master of Science in Computer Science or equivalent.
  
• Graduation in June 2027
  
• Fluency in Swedish, both written and spoken
  
• Fluency in English, both written and spoken
  
  

Purpose
In this thesis, the student(s) will investigate and answer the following:

• Local & Hybrid Deployment Feasibility: How can Gas Town/Gas City infrastructure be successfully deployed as a repeatable software factory running locally on developer machines and hybrid/private cloud infrastructure?
  
  
• Hardware-to-Task Optimization: How do different model tiers (CPU models, small GPU models, large 128GB local models, and external frontier models) perform across distinct orchestration workflows? Which capabilities are necessary for specific tasks?
  
  
• Cost, Performance, and Configurability: What are the quantitative tradeoffs regarding API cost, local compute utilization, latency, prompt configurability, and code accuracy when moving from a uniform model environment to a tiered hybrid architecture?
  
  
• Data Privacy and Feasibility: How well does the local/hybrid setup mitigate enterprise data governance risks compared to fully cloud-reliant configurations?
  

Methodology & Technologies

• Core Frameworks: Gas Town Open Source Orchestrator, Gas City SDK, and the Beads Data/Memory System.
  
• Hardware & Runtime Environment: Local execution environments (CPU/GPU workloads), local heavyweight execution (GB10 Grace Blackwell 128GB system), and hybrid/private cloud API gateways.
  
• Models and Protocols: Local open-weights models (e.g., Llama, Mistral, Qwen, DeepSeek-Coder) versus cloud-hosted frontier models (e.g., Claude, Gemini).
  
• Approach: Experimental software engineering. The student(s) will run standardized development scenarios across the different hardware configurations, measuring token burn rates, execution time, hardware throttling, and code validation success rates.
  
  Expected outcomes
  

• A working Proof of Concept (PoC) of an internal enterprise software factory powered by Gas City/Gas Town running in a local/hybrid hardware environment.
  
• An Enterprise Evaluation Matrix detailing cost, hardware requirements, configurability, and feasibility metrics across various local and cloud-based LLM architectures.
  
• A structural blueprint/guide illustrating the optimal mapping of specialized model sizes and hardware backends to specific agent lifecycle tasks.
  
  
  An Exciting Journey with Knightec Group
  Semcon and Knightec have joined forces as Knightec Group. Together, we are Northern Europe's leading strategic partner in product and digital service development. With a unique combination of cross-functional expertise and a holistic business understanding, we help our clients realize their strategies - from idea to complete solution.
  

Practical Information
This is a thesis position, located at our office in Sundsvall . Start date January or March 2027.

Please submit your application as soon as possible, but no later than 2026-11-30 . If you have any questions, you are welcome to contact johanna.edstrom@knightec.se . Note that due to GDPR, we only accept applications through our careers page.