Portfolio

Software Engineer — Python · TypeScript · AI/ML

I build production-shaped software — APIs, agents, and interfaces — with LLMs wired in as a first-class part of the system, not a bolt-on demo.

B.Sc. Computing Systems at Ulster University (predicted First Class), based in London. I like shipping things that behave correctly under real constraints — auth, concurrency, cost, and the edge cases users actually hit.

View projects Download CV
Portrait of Doniyor Mekhmonov

About

I write software that ships, not just notebooks. Finoks is an AR/AP agent with a FastAPI backend, a React UI, and tiered autonomy (AUTO / ESCALATE / HARD_BLOCK) so an LLM makes decisions without going rogue. My Telegram fitness assistant wires Claude into real workflows — Supabase, OAuth for Strava and Garmin, encrypted credentials, a production-shaped deploy. Underneath both is the same instinct: define the contract, handle the edge case, and keep the system explainable when it is wrong. The code is the proof — open a repo rather than take my word for it.

I am looking for a Software Engineer or AI/ML role where I can build end-to-end — backend, frontend, and the model integration in between. I bring Python and TypeScript fluency, hands-on experience wiring LLMs into real products (prompting, guardrails, autonomy tiers, tool use), and a data background (SQL, Python, R) that lets me reason about what is feeding a model, not just the API call. I currently work part-time at Aqua Shard alongside full-time study, which sharpened the pace and precision I bring to shipping under pressure.

Technical skills

Levels are honest signals, not decoration — tied to what I have actually shipped, not just studied.

Core stack / shipped in production

  • Python Advanced Finoks backend (FastAPI agent loop, SQLite, integrations) · Premier League analysis (Pandas, Jupyter, 15+ charts).
  • TypeScript & full-stack Shipped FastAPI · React/Vite · WebSockets · Telegram webhook, Supabase migrations, OAuth flows.
  • LLM & agent integration Shipped Claude tool use, prompt design, guardrails, and AUTO / ESCALATE / HARD_BLOCK autonomy tiers in production.

Data & infrastructure

  • SQL (MySQL) Intermediate Apple SQL project: schema, load, 30 business questions, CTEs, windows, views, stored procedures.
  • Auth & integrations Shipped OAuth (Strava, Garmin), encrypted credential storage, Docker deploys, Supabase/Postgres.
  • R · Power BI · Git Daily tools London bike analysis in R; dashboard coursework; versioned, reproducible work.

Competencies

  • API design & backend architecture
  • LLM / agent integration & guardrails
  • Auth & OAuth flows
  • SQL & data modelling
  • Testing & Docker deployment

Projects

Click a card to open the full write-up, charts where relevant, and the GitHub repository.

Apr 2026

Finoks — AR/AP agent (FastAPI + React)

Python · FastAPI · React · Vite · WebSocket · SQLite · Claude

Autonomous invoice workflow for UK SMEs: AML-aware screening, statutory interest logic, and AUTO / ESCALATE / HARD_BLOCK autonomy tiers with a live demo UI.

Open project → Telegram chat: bot reply with workout summary and a two-month training plan Apr–May 2026

Telegram fitness assistant (TypeScript)

TypeScript · Supabase · Claude · Strava OAuth · Docker

Workout sync, provider connect commands, encrypted credential patterns for Garmin, and Claude answers grounded in recent activity — production-shaped deploy story.

Open project → Aug 2023 data · Built 2025

London Santander Cycles (R)

R · RStudio · R Markdown · ggplot2

Trip duration, station gravity, weekday double peaks vs weekend leisure curve — TfL-style operational angles.

Open project → 2023/24–2025/26 · Built 2024–2025

Premier League performance (Python)

Python · Pandas · Matplotlib · Seaborn · Jupyter

Multi-season merge, league table from scratch, home advantage, correlation work, and a full chart set including margin wins and clean sheets.

Open project → 2022–2024 data · Built 2025

Apple global sales (MySQL)

MySQL · CTEs · window functions · views · stored procedures

11.5k transactions, 30 answered business questions, reusable views and five parameterised procedures.

Open project →

Upcoming

Agent evaluation harness

Starting next

A test suite for LLM agents — golden-path scenarios, guardrail regression tests, and scored transcripts — the same rigor Finoks's autonomy tiers need, packaged so any agent can be evaluated before a prompt or model change ships.

ML model — trained, served, monitored

Classic ML + deployment

A model trained from scratch, not just an LLM call — scikit-learn or PyTorch — wrapped in a FastAPI endpoint with versioning and basic latency/drift monitoring. Closing the loop from notebook to production service.

Work experience

Floor staff — Aqua Shard, London

Jun 2025 – Present | Part-time (~20 hrs/week)

  • High-standard guest service in a fine-dining, high-expectation environment.
  • Balanced with full-time study while maintaining a First Class average.
  • Composure, precision, and clear communication under pressure.

Education & awards

Ulster University — B.Sc. Computing Systems

Sep 2024 – May 2027 | London, England

Predicted classification: First Class

Key modules: Databases, Statistics, Machine Learning, Algorithms, Software Engineering

Dean's List

Academic excellence, Ulster University (2024–2025).

Contact me

Send a message — I usually reply within a day or two.