Sefineh T.Private profile: Upwork shows its earnings only to signed-in members, so we check it every few weeks, not daily. MRR here is a rough estimate (Upwork's six-month earnings ÷ months active). Last read Nov 5, 2026.

Founder of Lamied (https://lamied./com.et, 👈remove slash), an AI coding tool in the NVIDIA Inception program, the Google for Startups Cloud Program, and MongoDB for Startups. I design and build production AI systems end to end: the model layer, retrieval, agents, the backend services around them, and the infrastructure they run on. I make the architectural calls that decide whether an AI product stays fast, affordable, and correct as usage grows. On Upwork, I have delivered 10+ features across 20+ client contracts with a 100% contract completion rate. Systems I have built: - Native code intelligence: a C++ engine on tree-sitter that parses 14 programming languages into syntax trees and builds a cross-file symbol graph. It runs off the editor's typing path, so heavy analysis never blocks the user. - Agent architecture: a tool-calling agent that executes multi-step changes across entire repositories through 38 tools (Git, terminal, language-server queries, symbol-graph navigation), with user approval on every edit and Git checkpoints for rollback. - Retrieval at scale: a semantic codebase index on Voyage embeddings with content-hash incremental updates, so only changed files are re-embedded, plus a persistent vector store and cache-aware context assembly that sends the model only what a task needs. - Model abstraction: a provider-agnostic layer that runs NVIDIA Nemotron through NVIDIA NIM, OpenAI, Anthropic, Gemini, or local models behind one interface, so the product is never locked to one vendor. - Low-resource NLP: an open-source Amharic tokenizer (6,000+ PyPI downloads) using byte-pair encoding over decomposed Ge'ez characters with a Cython core. I re-engineered training to run 38x faster on 8x less memory with identical merges. - Production platforms: Tiragn, a teleradiology platform on Google Cloud built with FastAPI, React, PostgreSQL, and Docker. Engineering results for clients: - Eliminated 100+ security vulnerabilities and code-quality issues by enforcing SonarQube static analysis as CI quality gates. - Cut deployment overhead by about 60% by automating releases and versioning with GitHub Actions. - Reduced billing errors by 30% and made a dashboard load 50% faster for 200+ daily users at a US software company, through workflow redesign, code splitting, lazy loading, and request batching. What I can own for you: - Architecture for LLM products: model selection, retrieval design, agent tool design, context and cost control - RAG pipelines: ingestion, chunking, embeddings, vector search (Pinecone, Qdrant), and evaluation against your real queries - Fine-tuning with LoRA and NLP for specialized or low-resource languages - Backend and full stack: Python, C++, TypeScript, FastAPI, Django, Node.js, React, PostgreSQL, MongoDB - Infrastructure: AWS, Google Cloud, Docker, Terraform, CI/CD How I work: I start by defining what success means and how we will measure it, then deliver in milestones with working software at each step, tested and documented code, and a clean handover. Education: BSc in Software Engineering (AI stream), Addis Ababa University. Certifications in Retrieval-Augmented Generation (Duke University) and Natural Language Processing (IBM). Based in Addis Ababa (UTC+3), with full working-hour overlap with Europe and partial overlap with the US. Send me your project details, and I will reply with a proposed architecture, milestones, and timeline.