AI-Native Development
![Cover for AI-Native Development](/static/d1fceb5fa9c1548c03cf163e7589c9a1/6bdd7/cover.png)
Share on
Skills
- Neural Networks
- Lean Software Development
Stay focused on your core expertise while applying recent AI advancements. Boost your topline and cut expenses—without the need to research and navigate the constantly changing spectrum of new models, tools, and techniques. As your partner, we help design and build pragmatic AI-native solutions, usually falling into one of two categories: native conversational UX and manual task augmentation.
1. Native conversational UX
AI-native applications embody the overlap between traditional and conversational UX: a person communicates using natural language (text or voice), but the app responds with interactive components. The user then interacts with these components, and the app performs a meaningful action, such as: booking a flight, making a purchase, scheduling an event.
Implementation involves a few steps:
- Exposing API endpoints to LLMs as tools
- Applying a decision-making framework for the LLM to select a tool
- Building interactive components that wrap the responses of each “tool” and expose the requested functionality in a precise and contextualized manner
2. Augmenting manual tasks
Today, using LLMs to automate mundane manual tasks is most efficient when a human’s work is not replaced, but instead augmented by AI. The probabilistic nature of generative AI means the final result always needs to be checked by the responsible person, sometimes edited, and in rare cases, discarded. Therefore, effective design of this kind of automation smoothly integrates the “draft” result right into the work process.
Implementation includes these steps:
- Understanding the specialist’s workflow and where and when an AI-generated “draft” of the work makes most sense
- Rapid iterations on prompting and tools (the particular LLMs being used) to maximize the applicability of resulting “drafts” and ensure economic viability (taking into account the cost of different solutions) in the testing environment
- Production implementation and release
- Monitoring of quality metrics (the actual usage of “drafts” in final work)
Choosing Evil Martians
When seeking a trusted parter, it’s worth looking at a brief snapshot of the Evil Martians picture:
- 18 years of early-stage startup software consulting; we thrive building things nobody has built before
- 8 years applying machine learning and AI technologies in tech startups, including recommendation algorithms, image recognition, and building tools for machine learning teams
- A successful exit with application-level AI: CanFy, our internal Martian-built RAG startup, was acquired by Playbook
Sounds interesting? Contact us using the form below!