Engineering knowledge transfer at Google has a structural problem: the people with the answers are hard to find, harder to ask, and often unavailable when you need them. The cost of getting unblocked is social as much as technical.

Gemini TL was an AI tech lead assistant embedded in Google Chat, designed to reduce that friction at every stage — from helping an engineer formulate a better question, to surfacing the right expert, to facilitating the introduction with context and care.

Five Interaction Models

The system was designed around five distinct scenarios, each representing a different moment in the knowledge-transfer lifecycle:

• Technical Q&A with confidence calibration — adaptive clarification loops that improved answer quality before surfacing a response
• Right Question, Right Person (RQRP) — a structured question-formulation scaffold that prepared engineers to engage experts effectively, reducing the social cost of asking
• Proactive SME recommendation — on bug assignment, the system surfaced top 2–3 experts with match confidence scoring and justification
• AI-facilitated introductions — mediated connections with context, timing awareness, and drafted outreach
• Work packaging — on-demand assembly of engineering context (CLs, bugs) for knowledge sharing

Research & Validation

Conducted 12 one-hour UXR sessions with engineers across Core Engineering teams. Received extremely positive feedback across all tested flows. The RQRP model was the standout — engineers reported that the scaffolding process itself changed how they thought about their problem before they'd even reached the expert.

Before you can get a good answer, you need to be able to ask a good question. The system taught engineers how to ask.

Outcome

Following a team reorg, the project was deprioritized. The core interaction model — particularly the RQRP scaffold and the SME recommendation flow — was incorporated into a broader Google-wide rollout.

Narrative Position

Gemini TL sits at the exact hinge between Act 2 and Act 3 of this career. Act 2 was about designing knowledge systems — ways to embed expertise, share it, scaffold it, make it accessible. Act 3 is about AI systems — autonomous agents, decision delegation, human-in-the-loop design. Gemini TL was both simultaneously: an AI system designed to improve the human knowledge systems it was embedded in.