Hi, I'm Maya Kid 👋

Applied AI · Agent Engineering · Retrieval & Knowledge Graphs · Multimodal Systems

I work on the details that make AI dependable — from evidence-grounded answers to resilient agents and reliable data pipelines.

_{105 merged PRs to Utopia as of October 7, 2026 · Explore the contributions ↗}

---

🧭 About Me

I'm a developer with a background in mathematics, working at the intersection of AI agents, information retrieval, and real-world interaction. I enjoy taking a useful idea through the engineering needed to make it dependable: clear permissions, traceable evidence, reproducible evaluation, and human oversight.

• 🔭 Open-source focus: contributing to Utopia, an open-source enterprise world model, with work across retrieval, graph reasoning, agent conversations, and data connectors.
• 🔎 Retrieval & evaluation: hybrid lexical and semantic search, two-stage ranking, held-out evaluation, and preventing data leakage.
• 🤖 Agents & tooling: MCP, structured outputs, tool calling, permission-aware execution, and human approval flows.
• 🦾 Beyond the screen: multimodal sensing, depth cameras, IMUs, voice interaction, and connecting model suggestions to controlled hardware execution.

🛠️ Tech Stack & Tools

Systems · RAG · Knowledge graphs · Tool calling · Human-in-the-loop
Evaluation · Recall@K · NDCG · Precision · Latency · Ablation studies

---

✅ Selected Open-Source Contributions

Utopia · Contributor
Selected merged PRs, with the engineering problem behind each change.

Contribution	What changed
Evidence-grounded answers	Linked graph answers to source facts so readers can inspect the evidence. #968
Resilient agents	Added history budgeting and context-overflow recovery for long conversations. #973
Retrieval correctness	Stabilized tied search results and fixed filtering order that could drop valid matches. #1060 · #728
Efficient embeddings	Reused each turn's question embedding across mapping selection, retrieval, and graph tools. #978
Temporal reasoning	Completed transitive inference chains when a missing premise arrives later. #1051
Data integrity	Corrected Databricks and Snowflake partition reads; preserved comments in incremental GitHub sync. #1063 · #1062 · #1061

View all merged contributions →

🔬 Broader Engineering Work

Alongside open source, my work spans enterprise knowledge workflows, candidate retrieval and ranking, and multimodal sensing.

Area	Engineering focus
Agent workflows	Structured tools, explicit permissions, approval flows, traceable execution, and idempotent writes.
Retrieval & ranking	Hybrid candidate retrieval, two-stage ranking, and held-out evaluation with attention to leakage.
Multimodal robotics	Depth and inertial sensing, voice interaction, and clear boundaries between model suggestions and hardware actions.
Applied ML experiments	Low-cost spectral sensing, sample-level validation, matched ablations, and end-to-end demos.

_{Private work is described at a technical level; client identities, private datasets, and internal implementation details are omitted.}

---

🤝 Let's Build

I'm interested in collaborating on reliable agent tooling, retrieval and evaluation, knowledge systems, and multimodal applications.

If you're working on something in these areas, I'd be glad to compare notes, review an idea, or contribute a fix.

_{Curious about the model. Careful about the system.}