EcoSynth.ai

Inspiration

If there's one thing I learned in corporate sustainability, it's that being in an analyst role is tedious, and much of your job will probably entail trudging through boring manual tasks - scouring emissions databases, writing annual reports, and scrolling through treacherously long excel spreadsheets. It's really a shame knowing that while this occurs, our species continues to carelessly pump carbon into the atmosphere, tabling a discussion of the consequences to a later date.

Fortunately, in order to combat this foolishness, several of our governments have enacted stricter regulation such as the Corporate Sustainability Reporting Directive (CSRD), requiring the private industry to more rigorously take accountability on their impact towards the environment. Carbon accounting, life cycle assessments, and environmental product descriptions are all examples of sustainability reporting frameworks slowly being mandated for companies to conduct in order to assess their overall footprint on our planet, whether it be land use, manufacturing emissions, etc.

Unfortunately, these documents are extensive and data intensive, meaning that our poor sustainability analysts are seeing an exponential increase in work that needs to be done each quarter - it might not be possible to keep up.

What it does

This is where EcoSynth steps in. Utilizing recent AI agent developments, we now have the capability to build an analyst's best friend by simply having an autonomous AI gather data from any digital source, perform due diligence, all while abiding by international reporting standards.

EcoSynth uses AI to drastically cut down on manual labor for sustainability consultants by automatically searching through a variety of emissions factor databases (provided by the EPA, EEA, other international organizations, etc.) to accurately estimate, for example, the total GHG emissions released per the manufacturing of one iPhone.

How we built it

Using open source LLM models from Together AI, and an AI agent framework like LangGraph/LangChain, we were able to put together a research team of autonomous bots with API calling capabilities that would scour various reputable sources for relevant research. With this, our AI research team was able to communicate with each other, delegating tasks to ultimately find accurate estimates for carbon emissions from manufacturing processes, transportation of goods, and electricity usage with high granularity to geographical location. All of this was packaged into a simple web application using the Reflex front-end framework.

Challenges we ran into

Sometimes working with language models can be tough because they can produce inconsistent results and behave erratically. This was a major challenge we faced head on through trial and error while testing our application. Fortunately, with some clever LLM prompting techniques and adjustments to the architecture of the multi-agent network we had built, it is possible to mitigate most of these issues.

Accomplishments that we're proud of

When we gave our AI research team a simple bill of materials for a construction process, we were able to have our AI agents quickly gather emissions factors, and work together to generate a basic sustainability report on GHG emissions in a matter of seconds and under 5 cents.

What we learned

We learned a lot about the power of autonomous agents and how the application of such technology could completely disrupt things like research, data analysis, content creation, and so much more. It seems pretty apparent that the future of many white collar jobs will be either completely enhanced by the elimination of the need to perform tedious tasks, or completely eliminated altogether.

What's next for EcoSynth.ai

I'm confident that the technology behind EcoSynth could be scaled to an extent in which it could outperform a human analyst in both the speed and cost benchmarks by orders of magnitude. I'm also very certain that the capabilities of EcoSynth goes beyond just sustainability reporting - the future of any digital task that is tedious will likely be automated as AI technologies become increasingly more effective at multi-model use cases. I hope to continue building off of EcoSynth's technology and begin trialing versions of this product with consultants at various firms - this is an interesting development to pursue.

Built With

• javascript
• langchain
• langgraph
• llama
• openai
• python
• reflex
• together.ai