I build reliable agentic systems, evaluation-driven RAG pipelines, and production AI products—taking them from ambiguous requirements through deployment, observability, and measurable iteration.📍 Based in Addis Ababa (EAT / UTC+3) — Open to remote roles worldwide and relocation for onsite Applied AI opportunities.
30+
10K+
5yr+
I'm an Applied AI Engineer specializing in agentic systems, evaluation, RAG, and full-stack product delivery. I connect modern models to dependable APIs and thoughtful user experiences that solve concrete business problems.
Whether I'm shaping an ambiguous requirement, implementing an MCP integration, designing a retrieval pipeline, or shipping the final interface, I engineer for reliability, measurable quality, and real-world use.
“Useful AI is more than a model call—it is a system that can be measured, trusted, and improved.”
What: Defining user outcomes, representative eval sets, quality thresholds, and failure categories before scaling a solution.
Why: Turns subjective AI demos into measurable product behavior and clear release decisions.
What: Combining agents, retrieval, structured generation, and model integrations with dependable Python and TypeScript services.
Why: Keeps model behavior observable, bounded, and useful under real product constraints.
What: Shipping the interface, APIs, data flows, cloud infrastructure, monitoring, and feedback loops around the AI capability.
Why: Moves the work from a promising prototype to a product people can adopt and improve.
AI, web, and mobile products delivered
People served by production systems
Production defects reduced through typed workflows
Team productivity gained from agent workflows
Applied AI capabilities
A delivery stack organized around the work: evaluate model behavior, connect retrieval and tools, build dependable services, ship the product surface, and observe it in production.
33 production capabilities across the AI lifecycle
I turn ambiguous product goals into measurable AI behavior—aligning with users, designing the system, validating model quality, and shipping the complete production experience.
I translate user needs and business constraints into a testable AI behavior, an evaluation set, and a clear definition of useful before implementation expands.
I combine strict Pydantic and TypeScript contracts with targeted evaluations, observable workflows, and explicit failure handling so AI behavior can be measured and improved.
I connect model and retrieval behavior to backend services, data flows, interfaces, deployment, and production feedback so the AI capability survives contact with users.
Selected Applied AI Work
Case studies showing how I turn models, data, evaluations, and full-stack engineering into production systems with measurable behavior.
FastAPI sales automation engine for synthetic-prospect outreach, combining inbound email/SMS webhooks, HubSpot write-back, Cal.com booking, enrichment signals, Langfuse traces, and tau2-bench evaluation.
“Built the sales agent with sink-routed outbound safety and sealed-eval guardrails before any live-channel workflow.”
Sales-domain benchmark and trained critic layer for Tenacious-style B2B outreach, with contamination-aware task generation, deterministic scoring checks, preference data, and a small LoRA judge.
“Turned prospect-facing sales failures into a measurable benchmark and judge gate instead of trusting generated drafts by default.”
PDF triage and extraction pipeline that detects document origin, layout, and domain, escalates extraction strategies by confidence, builds PageIndex trees, and answers with provenance chains.
“Converted messy PDFs into auditable structured facts with confidence gates, retrieval indexes, and source-cited answers.”
LangGraph forensic codebase evaluator using a Digital Courtroom protocol: evidence-gathering detectives, three judge personas, deterministic conflict resolution, and hallucination policing.
“Made repository evaluation adversarial and evidence-bound instead of letting a single judge model grade from vibes.”
Engineered a modular, production-grade backend framework designed specifically to power AI workspaces. Implements robust token caching, graceful rate-limit handling, circuit breakers, and WebSockets for real-time state streaming.
“High-performance, async-first engine for long-running agentic processes and real-time streaming.”
A fast mobile trading app for crypto and prediction markets with AI-assisted insights, live market data, interactive charts, paper trading, risk controls, strategy review, and performance monitoring.
“Built an Expo/React Native trading client that balances fast multi-market access with AI decision support, simulation, persistent state, secure storage, and disciplined risk tooling.”
View all 26 projects and case studies
Reproducible explanations of the evaluation, alignment, and inference decisions behind production LLM systems—not a glossary of AI terminology.
A production case study in turning domain failures into a measurable benchmark: contamination-aware task generation, deterministic checks, human grading, preference data, and a trained judge.
A first-principles explanation of paired versus unpaired bootstrap designs for LLM evaluation, backed by Python simulation and an explicit variance analysis.
Objective functions, length bias, reference-free optimization, gradient behavior, and VRAM tradeoffs for preference tuning.
Why prompt instructions are weaker than constrained decoding, schema enforcement, and token-level generation controls.
A systems-level explanation of adapter merging, parameter updates, runtime graph shape, and where LoRA inference overhead actually appears.
The research layer of my Applied AI practice: methods I can explain, reproduce, test, and carry into production systems.
View DEV profileVerified Delivery
Delivered a complete social music party-game product engagement from implementation through client handoff.
Verified client engagement
January–June 2026
Project
Social Music Party Game
Impact
Verified 5.0 Upwork rating
30+
projects shipped
10K+
users reached
65%
bug reduction
Production Experience
A progression from product and platform engineering into production AI—combining model integration, backend reliability, and end-to-end delivery.
Tell me about the users, workflow, data, and quality bar. I can help frame the problem, build the system, and create the evaluation loop around it.
Share the problem, current system, and what a successful outcome looks like.