Ashkan Lotfipoor

Now: Data Scientist at Cycle Confident

Ashkan Lotfipoor.

AI engineer who ships. I fine-tune and deploy LLMs, build the pipelines around them, and ship the products that carry them to users.

Experience

Data Scientist, KTP Associate

Cycle Confident & University of Surrey · Feb 2024 – now · London

I build The Hub, Cycle Confident's analytics and operations platform, and the AI inside it.

  • Fine-tuned and deployed deepseek-llm-7b-chat on 2M+ domain documents, improving query accuracy by 40% over a general model
  • Built the platform in React, Django and PostgreSQL for 50+ staff and 200+ external clients, cutting data retrieval time by 65%
  • ETL pipelines processing 500K+ records a day on Docker and Railway, at 30% lower infrastructure cost
  • Published the LLM research in the Journal of Computer Information Systems (2026)
Django · PostgreSQL · React · PyTorch

Freelance Mobile Developer

Radzi · Oct 2025 – now

An active travel app, shipped to the iOS App Store from design through App Review.

  • GPS route tracking, group goals and leaderboards
  • A street-level feedback layer that gives local councils walking and cycling infrastructure data

Researcher (part-time)

Siemens AG, research project · May – Nov 2022 · London

CNN-based fault detection for high-speed train bearings: 92% accuracy on unlabelled acoustic data, published in Sensors.

Research Assistant

Heriot-Watt University · Sep 2021 – Feb 2023 · Edinburgh

River-flow forecasting with gradient boosting, 25% more accurate than baseline methods, and sensor data from 10K+ UK homes for energy-efficiency analysis.

Senior Data Analyst

Samsung · Jun 2017 – Sep 2019 · Tehran

Automated reporting with Power BI dashboards tracking 50+ KPIs, cutting manual analysis time by 40%.

Ventures

Zynthia

Early access

Founder & Engineer · Apr 2026 – now

A living systematic literature review platform: saved searches are monitored, new papers arrive on their own, and the review never goes stale.

  • AI screening that suggests include or exclude with a Relativity Index score and verbatim evidence quotes; the researcher decides
  • Every decision is logged and reversible, with contradiction flags and one-click PRISMA export
  • 150+ researchers on the waitlist ahead of the October 2026 launch
zynthia.ai(opens in a new tab)Read the write-upNext.js · Django · Gemini · PostgreSQL

Veticity

Pilot

Co-founder · 2026

Pre-submission review for business and management researchers targeting top journals: AI plus a matched human expert check a manuscript against its named target journal.

  • A manuscript workspace where findings are anchored to the text and each revision is re-checked
  • Journal fit judged from fetched facts, each stored with its source and date
veticity.com(opens in a new tab)Next.js · Django · PostgreSQL

Research

PhD, deep learning for energy forecasting

Heriot-Watt University · James Watt Scholarship · Oct 2019 – Mar 2024

Thesis: Enhancing Energy Demand Forecasting and Data Imputation Using Deep Learning. A Transformer plus K-means imputation model and a CNN-LSTM forecasting framework with Bayesian optimisation, feature decomposition and bias correction.

Publications

  1. 2026Large Language Models to Analyze Business Intelligence User NarrativesLotfipoor, Fakhimi, Hagen-Zanker & Showell · Journal of Computer Information Systems
  2. 2024Deep neural network with empirical mode decomposition and Bayesian optimisation for residential load forecastingLotfipoor, Patidar & Jenkins · Expert Systems with Applications
  3. 2023Transformer network for data imputation in electricity demand dataLotfipoor, Patidar & Jenkins · Energy and Buildings
  4. 2023Intelligent bearing fault diagnosis based on feature fusion of one-dimensional dilated CNN and multi-domain signal processingDong & Lotfipoor · Sensors
  5. 2022Missing data imputation for community energy demand modellingPatidar, Jenkins, Peacock & Lotfipoor · IBPSA-Scotland uSIM 2022
  6. 2020Short-term forecasting of residential electricity demand using CNN-LSTMLotfipoor, Patidar & Jenkins · IBPSA-Scotland uSIM 2020
All citations on Google Scholar

About me

I started in research, with a PhD in deep learning for time-series forecasting, and moved into building products. Today I'm the data scientist on a Knowledge Transfer Partnership between Cycle Confident and the University of Surrey, where I build The Hub. On the side I'm building Zynthia and Veticity, and I shipped Radzi to the App Store.

Outside work I'm usually reading two or three books at once, out on a hiking trail in the English countryside, or making a mess of the kitchen with a new recipe. I'm also pretty good at Hollow Knight: Silksong.

Commits, day by day

568 contributions in 2026, across 98 active days. Longest streak 7 days; busiest week from 7 Jun with 84.

Hover a day to read it.
github.com/alotfipoor

Tools I use

PyTorchTransformersLoRA fine-tuningRAGLangGraphTime seriesDjangoNext.jsReact NativePostgreSQLDockerRailway

Writing

All posts