Yusuf Nuru Yesuf

• Deployed and maintained production applications and databases using CI/CD pipelines with AWS ECR and Portainer, reducing manual deployment effort by 10%.
• Optimized query performance across multiple pages, reducing latency by up to 50%, significantly improving application responsiveness.
• Developed a multi-agent agentic platform for 3-way matching within an ERP system, building a multi-layer agent system using Pydantic AI as a dedicated service, reducing human verification effort by 90%.
• Collaborated with developers to build and deliver a full ERP management system for a scaffolding company, covering inventory management, work order processing, purchase requests, and end-to-end scaffolding workflow management.
• Identified and reported critical security vulnerabilities (token stored in local storage) and business logic issues during QA/QC, strengthening platform security and reliability.
• Automated local database seeding, cutting environment setup time by 10% and simplifying onboarding for new developers.

• Authored and maintained 5+ pages of technical documentation and debugging guidelines, improving onboarding efficiency for future interns and developers.
• Collaborated with developers to enhance platform performance by optimizing component rendering.

A self-hostable observability & monitoring platform for AI agents and LLM applications, covering trace ingestion, agent-run analytics, and multi-channel alerting with multi-organization RBAC. Currently in active development.

• Designed and built a multi-tenant SaaS from scratch to trace LLM calls, API events, infra metrics, and agent runs, solo across a Go backend, React dashboard, and two client SDKs.
• Built a high-throughput ingest pipeline: a generic batch buffer flushing via Postgres COPY into TimescaleDB hypertables (time-partitioned, per-tenant segmented), with per-IP and per-API-key rate limiting.
• Implemented a stateful alerting engine: a rule evaluator with a pending→firing→recovering→resolved state machine plus a transactional-outbox delivery worker (retry/backoff) fanning out to email, Slack, webhook, and PagerDuty.
• Shipped agent loop detection via SHA-256 fingerprinting of tool-call inputs, surfaced in the API and dashboard.
• Built full auth & RBAC: kw_-prefixed API keys (SHA-256 hashed), cookie sessions, Google/GitHub OAuth, email verification, and org/project roles (owner, admin, member).
• Defined a single ingest wire protocol in the Go core and authored two SDKs against it — Python (sync + async, adapters for Anthropic, OpenAI, pydantic-ai) and TypeScript (Vercel AI SDK adapter) — both with context-manager runs, decorators, and typed errors.
• Delivered a React dashboard with agent performance, runs, tool analytics, and team/project settings views.

A university marketplace platform for students to buy and sell academic materials, textbooks, and campus essentials, built as a final year project.

• Built a full-stack solution with a Vue.js frontend and Express + PostgreSQL backend.
• Implemented user authentication, product listings, in-app messaging, and responsive design.
• Conducted end-to-end testing to deliver a fully functional, user-friendly platform.