Hands-on R&D Process
How we operate Research & Development tasks at AIDo — helping members master emerging technologies with clear direction, avoid rabbit holes, and build the incubator's shared knowledge vault.
“Research at AIDo is never about reading passive theory. Research means building a tangible, working Mini PoC.”
Task Initiation — Top-down & Bottom-up
R&D tasks at AIDo originate from two open streams:
Mentor Delegated
Solving specific partner challenges. Mentors must provide a concise, structured Research Brief.
Member Proposed
Any member can initiate an R&D task from a personal curiosity, provided it leverages AI and yields new insight. Your spark might pivot the team's trajectory.
Research Brief Structure (Mandatory)
- Context (The “Why”) — Which project bottleneck does this solve, or what is the practical utility of this individual idea?
- Core Constraints — API cost budget, data constraints, Local vs. Cloud requirements. For partner data, security constraints are absolute.
- Deliverables Definition — Explicitly define expected outputs: executable script, comparative model benchmarks, etc.
Time-boxing & The Red Flag Rule
- Time-box Rule — Every R&D task is boxed within 1 to 3 working days, preventing members from spiraling down unproductive rabbit holes.
- Red Flag Rule — If 50% of the allocated time elapses and the code still fails to run, it is mandatory to raise a red flag in the group channel for collaborative debugging. Never struggle in isolation.
- “Ask for Prompting, Not Code” — Mentors guide mental models and prompt engineering strategies rather than handwriting code for you, strengthening your problem-solving muscle.
Acceptance — Show, Don’t Tell
An R&D review is never a dry Word document. It is an energetic 15–20 minute live showcase, via video call or in-person:
- Live Demo — Run real code, display genuine output on screen. Share the exact friction you wrestled through with prompts (your "scars") to arrive here.
- Limit Discovery — Answer candidly: under which edge cases does this model or tool fail? Discovering AI weaknesses is far more valuable than showcasing happy paths.
- Architecture Critique — The whole team asks probing questions on scalability, security, and cost optimization to elevate system thinking.
Packaging & Cross-pollination
Whether an R&D task succeeds or fails (proving a technology is unfeasible), it must conclude with two steps:
Step 1 — Standardize to Knowledge Base
Mandatory for preserving institutional knowledge: the “Golden” Prompt, core code snippets, and warnings against counter-intuitive pitfalls encountered.
Step 2 — Direct Sharing & Ignition
Knowledge stored solely on Notion remains cold. Once packaged, bring the story to a team coffee or host a flash Live Audio session on Discord. Use your authentic energy and pride to stimulate others — real passion spreads human-to-human, not through static pages.