Agentic AI systems
Tool-calling assistants, RAG pipelines, policy guardrails and eval harnesses wired into CI.
Available for the next challenge
Fullstack Engineer · AI Platform & Developer Tooling
Tania Riaz
TypeScript · Node.js · Python · AWS
I build products that feel intelligent—agentic AI systems, config-driven platforms, and the developer tooling that makes them safe to ship.
5+Years shipping
5Certifications
3Languages
Human-led · AI-enabled
Profile
From demo to product.
Fullstack engineer with 5+ years shipping production web platforms, now working on agentic AI systems and the tooling that makes them safe to ship. I design config-driven platform architectures with enforced package boundaries, build tool-calling agents with policy guardrails, and put golden-set eval harnesses into CI as release gates — the unglamorous work of moving AI features from demo to product.
I both use and extend agentic development tooling day to day: custom Claude Code skills, versioned agent conventions, hash-pinned skill registries. Comfortable with ambiguity in early-stage products, and opinionated about developer experience — the best internal platform is the one nobody has to read a wiki to use.
Focus
Where I spend my depth.
Developer platform architecture
Config-driven multi-tenancy, enforced package boundaries and monorepo workflows.
Evaluation & safety gating
Golden-set runners, deterministic graders and release gates that keep AI features from shipping half-baked.
Developer experience
Internal consoles, custom agent skills and conventions that reduce wiki-reading to near zero.
Selected Platform & AI Work
Systems with a pulse.
01Featured
Freelance · architect & lead developer — 2026
Grounded AI Assistant Platform
Multi-tenant, config-driven platform where brand, domain, geographic scope and assistant persona live in a site-config package; the content schema is a portable package with an enforced no-React boundary — so standing up a second site is an instantiation, not a rewrite. The agent is a tool-calling assistant over RAG with search, matching and booking tools; safety policy is implemented as unit-tested modules, not prompt text alone. A generic golden-set runner plus deterministic graders across seven compliance categories is wired into CI as the single launch gate. Operator UX includes an internal console for agent testing, model and persona settings, and conversation review.
AnthropicVoyage embeddingsRAGTypeScriptCI eval gatesconfig-driven architecture
02
Client engagement · AI service owner — 2025–2026
Årshandlingar
FastAPI service on AWS Bedrock extracting eleven structured checkpoints from Swedish association protocols in PDF, DOCX and image form. Drew the architectural line the product rests on: extraction is probabilistic, adjudication is deterministic — the AI service returns structured data and never makes the approve-or-escalate call. Model choice stays an env var, so promoting a model is a config flip. Includes PII redaction on request and response paths, concurrency-bounded model calls, audit events over a shared secret, and a benchmark harness with versioned result sets and an HTML report. Shipped to ECS Fargate in eu-north-1 by GitHub Actions → ECR rolling deploy.
FastAPIAWS BedrockClaude Sonnet / OpusECS FargatePII redactionGitHub Actions
03
Ongoing, across every repo
Agent Tooling & Developer Experience
Author custom Claude Code skills — including a QA-sweep skill that runs unit, regression and end-to-end suites across an app and reports back on maintainability. Maintain AGENTS.md / CLAUDE.md conventions that pin coding agents to exact versioned framework docs, which stops them confidently writing against APIs that no longer exist. Keep a hash-pinned skill registry so agent tooling stays reproducible across repos and machines.
Claude Code skillsAGENTS.md conventionsskills-lock.jsonagentic DX
Technical Skills
Depth where it matters. Range when it counts.
AI & Agents
Anthropic Claude API, AWS Bedrock (Converse, inference profiles), multi-provider routing, tool calling, RAG & embeddings, eval harnesses & golden sets, prompt versioning, prompt-injection safeguards, PII redaction, IBM watsonx, Claude Code skill authoring
Platform & Architecture
Monorepos (pnpm workspaces), package boundaries, config-driven multi-tenancy, tenant isolation, API design, GraphQL (Apollo, Yoga, Pothos), REST, Prisma, PostgreSQL
Cloud & DevOps
AWS (Bedrock, ECS Fargate, ECR, S3, Amplify), Azure, Docker, Kubernetes, GitHub Actions, Vercel, Supabase, self-hosted Jitsi
Languages
TypeScript, JavaScript (ES6+), Python, C#, SQL, HTML5, CSS3
Frontend & DX
React, Next.js, Expo / React Native, component-driven development, Tailwind, Figma & design systems
Testing
Playwright (E2E), Vitest, Jest, Testing Library
Ways of Working
Agile / Scrum, SAFe, code review, technical writing & knowledge sharing, structured incident handoffs
AWS Cloud PractitionerSAFe 6 Scrum MasterIBM watsonx Essentialswatsonx.ai Technical Saleswatsonx.data Technical Sales
Work Experience
Built in the real world.
From polished interfaces to distributed services and AI safety gates, I follow the problem across the stack.
Apr 2025 – Present
Fullstack Consultant
Qualipro AB · Malmö, Sweden
- Breadth: ~1,000 commits across ten repositories of the AiQlick talent platform — Next.js frontend, Node.js GraphQL backend, Python matching services, a support backend, and self-hosted Jitsi video infrastructure. Third-largest contributor on a team of three core developers.
- AI capability work: built the AWS Bedrock Claude provider for the document-extraction service behind a pluggable provider interface. It is multimodal and authenticates through the default AWS credential chain, requiring no new organisation secret. Wrote the team’s documentation on provider routing and prompt-injection safeguards.
- Meeting insights: worked on LLM transcript analysis for the video platform; gated raw export behind super-admin after identifying leaked prompt scaffolding, and added disclosure that recurring rooms are analysed over the full accumulated transcript.
- Product surfaces: major contributor to a config-driven card and block rendering system with field-visibility control and dual table/card views. Owned resume parsing, connections, pipeline and onboarding flows, plus much of the typed GraphQL operations layer.
- Cross-stack debugging: trace front-end symptoms into backend and schema, then write structured handoffs — root cause, fix, blast radius. Surfaced everything from an RBAC design flaw to a latent IDOR.
Jul 2022 – Feb 2025
System Developer
Atea Sverige AB · Malmö, Sweden
- Built Kim, an AI study and career counselor on IBM watsonx with an Unreal Engine 5 avatar — taken from concept to a product showcased on Swedish national television (SVT).
- Developed front-end and backend features in JavaScript, React and Node.js across several enterprise products.
- Enterprise integrations: Azure AD, Intune and GSuite administration tooling — internal tools for IT teams, not end customers.
- Worked directly with IBM on AI solutions; recognised for communication and cross-team collaboration. Agile delivery with Scrum, Jira and Azure DevOps; mentored junior developers.
Jan 2021 – Jul 2022
Developer (Freelance)
WEBBMI AB · Remote
- Responsive, cross-browser front ends in HTML5, CSS3 and React; REST endpoints in Node.js with MySQL.
- Delivered multiple client projects end to end, from requirements gathering through to deployment.
Education
BSc International Business
Lund University
2021–2024
Full Stack Web Development
Foo Coding, Malmö
2022
BSc Software Engineering
COMSATS University Islamabad
2021
Languages
English
Fluent
Swedish
Beginner
Urdu
Native
Projects & Community
Side work and shared learning.
Solo project — 2026
Mellaling
Cross-platform Swedish-learning app in Expo, React 19 and TypeScript — roughly 43k lines, nine Supabase edge functions and twenty-one database migrations, shipped to web via EAS.
ExpoReact 19TypeScriptSupabaseEAS
Volunteer — Feb 2024 – Feb 2025
AI Teacher · ReDI School of Digital Integration
Co-taught AI Fundamentals — machine learning and NLP on Azure — to career-changers and newcomers to Sweden.
Azure AI / NLPTeachingCommunity
Contact