Balancing Innovation, Trust, and Environmental Responsibility
At Aveni Labs, we’re passionate about the positive role AI can play in financial services. From expanding accessibility for underserved populations to supporting advisers with compliance, the benefits are significant.
But we also recognise the potential negative impacts of scaling large AI systems. That’s why, in developing FinLLM, our financial services-specialised large language model, we actively embed both human value alignment and AI trustworthiness guidelines. This ensures we remain not only innovative but also responsible, ethical, and compliant with regulation.
Our Commitment to People and Planet
AI does not exist in a vacuum. Every model has a real-world footprint – on people, on regulatory systems, and on the environment. At Aveni Labs, we are committed to minimising that footprint while maximising the benefits AI can bring.
- Remote-first operations: By operating remotely, we drastically reduce commuting-related emissions across our team.
- Responsible AI development: We consider environmental impact at every stage of model training, evaluation, and deployment.
We want to create systems that work for both people and the planet. Not systems that work at the expense of one or the other.
The Challenge of Training Large Models: Resource Consumption
It’s no secret that datacentres consume vast amounts of electricity and water to power and cool modern AI workloads. Even though larger models tend to deliver higher performance, they also consume more natural resources.
Our approach is pragmatic and efficient:
- Model size flexibility: We are building a suite of models of different sizes, using smaller models for simpler tasks and reserving larger models only for complex ones.
- Distillation and pruning: These techniques reduce parameter size without sacrificing performance, ensuring we don’t “use a crane to lift a pencil”.
- Efficient infrastructure choices: We monitor our cloud usage, optimise training configurations, and account for the Power Usage Effectiveness (PUE) of the datacentres we partner with.
At Aveni Labs, we build FinLLM for financial services with a dual commitment: delivering powerful models and ensuring they remain responsible, value-driven, compliant and sustainable
Parameter-Efficient Fine-Tuning (PEFT): Faster, Smarter, Greener
For enterprise financial services firms, agility is everything. Markets shift, regulations change, and customer expectations evolve. But training and retraining large-scale models is resource-heavy and expensive. That’s where Parameter-Efficient Fine-Tuning (PEFT) comes in.
By using methods like LoRA (Low-Rank Adaptation), we fine-tune only a fraction of the model’s parameters instead of the full set. This has two major benefits:
- Practical agility for FS firms: It allows us to rapidly experiment and adapt models to reflect new regulatory guidance or client needs without incurring the huge environmental and financial costs of full-scale retraining.
- Sustainability without compromise: Lightweight fine-tuning not only speeds up development but also significantly reduces the energy footprint of every iteration.
For financial institutions under increasing scrutiny to demonstrate ESG accountability, PEFT provides a tangible way to ensure that the AI systems they adopt are both high-performing and aligned with sustainability goals.
Measuring Our Impact: Transparency for Trust
Numbers matter, especially in financial services where accountability is paramount. That’s why we calculate and share the environmental footprint of our models.
When training our 7B parameter model, we recorded:
- Total power consumption: 247.55 MWh
- Equivalent carbon emissions: 130.93 tCO₂
For context, this is on par with other industry-leading models such as OLMo, which reported 239 MWh of energy for pretraining its 7B variant.
We are sharing this because transparency builds trust. Just as financial institutions must evidence compliance to regulators and clients, we believe AI providers must be clear about the true cost of innovation. By setting these benchmarks and holding ourselves accountable, we give enterprise clients the confidence that adopting FinLLM aligns with both their business needs and sustainability commitments.
Looking Ahead
Our mission is to deliver industry-leading AI for financial services that is trustworthy, ethical, and sustainable. This means carefully balancing performance with environmental responsibility at every stage.
We will continue to invest in techniques like parameter-efficient training, smart infrastructure choices, and transparent reporting to ensure our innovations serve both financial services professionals and the planet we all share.
