/4 Adding Flexibility – Supporting Multiple LLM Models (Solidsight v8)
I
Former analyst with expertise in data, forecasting, and resource modeling, transitioned to cybersecurity over the past 4 years (as of May 2024). Passionate about security and problem-solving, utilising skills in data and analysis, for cybersecurity challenges.
Experience:
Extensive background in data analytics, forecasting, and predictive modelling. Experience with platforms like Bugcrowd, Intigriti, and HackerOne. Transitioned to Web3 cybersecurity with Immunefi, exploring smart contract vulnerabilities.
Spoken languages:
English (Native, British), Arabic (Fus-ha)
Expanding the Tool's Capabilities
Smart contract analyses can vary greatly in complexity and purpose. Realising this, I added support for multiple LLM models, allowing users to tailor analyses according to specific needs. This flexibility transformed Solidsight into a much more versatile tool.
Multi-model Support
I introduced a simple yet effective model-selection prompt using Rich:
pythonCopyEditanalysis_model = Prompt.ask(
"Select your analysis model",
choices=["deepseek-r1", "qwen2.5-coder:3b", "gemma3:4b"],
default="deepseek-r1"
)
Challenges in Implementation
Different models required slight prompt adjustments due to their varied interpretations and output formats. It took careful prompt engineering and extensive testing to achieve consistency across models.
Lessons Learned
The value of modular design became clear—flexibility allowed Solidsight to adapt easily to varying analysis requirements.
We will stop here, and I’ll see you on the next one.
pxng0lin.
