ehsansadeghi.me
09 entries

Professional experiences

Where I've shipped AI, and what I owned at each stop.

§ Work
Jul 2026 - present
Manchester

Senior Applied AI Specialist

@ Innovation Lab · Vypr Validation ↗

Continuing the Innovation Lab's mission of prototyping and validating new ideas, now with a more technical, AI-first focus and more advanced methodologies. The work centres on building deeper layers of consumer and market intelligence on top of Vypr's behavioural-science platform.

  • Started Signals, an added layer of intelligence that helps customers understand their market and the behaviour of their potential opportunities: sizing up competitors and reading trends to tell whether a category is rising, peaking or saturated.
  • Running a pure R&D project on digital twins of consumers, forecasting how a person would answer survey questions from their past behaviour, demographics and everything else we know about them, using agentic AI, generative AI and large language models.
Apr 2024 - Jul 2026
Manchester

AI Product Manager

@ Innovation Lab · Vypr Validation ↗

Vypr is a market research platform based on behavioural science, where brands test concepts, names, packaging, pricing and messaging with real consumers. I work in Vypr's Innovation Lab, the R&D team that prototypes and validates ideas before engineering takes them on.

  • Took an AI-moderated qualitative research tool from PRD to working prototype: text and voice interviews in 8+ markets and languages, with transcript chat. PoC validated through customer demos with strong positive feedback, and engineering committed to productionising it inside Vypr Predict.
  • Built the AI data quality system used daily by Vypr's Community team. Over 50 heuristic, ML and generative AI checks process millions of responses and remove thousands of fraudulent panellists each month. Now the team's main tool, used daily with confidence, with 99.4% of flags approved on manual human review.
  • Led the synthetic respondent R&D project, simulating survey responses from twelve years of anonymised consumer data. Built AI agents proven to answer surveys with precision close to real panel data.
  • Raised a customer-facing AI recommender from around 70% to 97% accuracy by building an internal evaluation platform with golden datasets, LLM-as-judge scoring, and cost and latency tracking. Now owned by product for ongoing AI quality maintenance.
  • Originated and prototyped the AI features now live in Vypr Predict, including summarisation, multilingual translation, sentiment analysis, and an agent that recommends surveys from brief customer input using internal knowledge.
  • Built Vypr's AI capability from zero. Ran multiple hackathons across around 25 people, introduced n8n, and established agentic coding practice. The Innovation Lab was the first team using Claude Code as its primary development tool.
Jul 2023 - Apr 2024
London

Co-founder & Product Manager

@ Teraformer

Teraformer was a generative AI startup that let you talk to anything: books, documents, memories, people or events, all turned into conversational digital twins.

  • Co-founded Teraformer and shaped product strategy, roadmap and architecture for an LLM-powered platform that lets users converse with the digital twin of a chosen subject.
  • Designed and built the RAG pipeline behind every digital twin. Combined hybrid search, GraphRAG for multi-hop questions and cross-encoder reranking for answer quality, with domain-tuned chunking and embeddings. Solved the latency and cost problems of running RAG at scale, the binding constraint at the time.
  • Shipped multiple working digital twins to real users across different domains, from personal memory archives to museum collections.
Mar 2022 - Jul 2023
London

NLP Product Owner

@ VoxSmart ↗

VoxSmart turns voice and text from financial markets into structured signals on quotes, trades and client intent, used by revenue, research and compliance teams.

  • Owned the post-acquisition integration of GreenKey's NLP technology into VoxSmart, managing the knowledge transfer between the two organisations and shaping the combined product roadmap with the AI team.
  • Took the inherited ASR and NLP components from research quality to production, improving accuracy and latency to the level required for live market deployment.
  • Researched and prototyped new technologies into working proofs of concept, validating what was viable before committing engineering effort.
  • Worked with Engineering on the cloud SaaS architecture, including scalable REST APIs and the data stores underpinning the platform.
Jul 2021 - Dec 2021
Bristol

Technical Product Manager (Contract)

@ Yantra Digital ↗

AI and machine-vision worker-safety platform for Shell Corporation, monitoring personnel in high-risk environments such as oil refineries.

  • Trained two computer-vision models on AWS to 96% precision in production, undercutting the cloud-AI alternatives on cost.
  • Owned product discovery and the roadmap between Shell's business stakeholders and engineering; set up the product team's first OKRs, KPIs and Agile (Scrum) cadence, with sprint-estimation error under 15%.
Nov 2019 - Jul 2021
Manchester

Co-founder & Product Manager

@ Safe Stream

Vision Inventory, a B2B computer-vision SaaS for grocery shelf monitoring. Multi-object detection (tens to hundreds of products per shelf image) plus fine-grained recognition against a ~10,000-SKU catalogue.

  • Built model-assisted annotation tooling: the model proposed candidate labels for each detected object, so annotators corrected rather than labelled from scratch, cutting both time and error. An early model-assisted-labelling (data-engine) approach, built 2019-21.
  • Led product discovery and a data-science and engineering team through customer interviews and market analysis; ~150K objects annotated across 2.5K shelf images for training, with end-to-end deployment on AWS.
  • Built and demoed the platform to retailers; commercial traction stalled in the COVID-era retail downturn.
Dec 2018 - Nov 2019
Tehran

Product Manager, AI Trading Platform

@ Tosan Ofogh ↗

A FinTech platform that traded equities automatically from AI signals, paired with a dashboard for tracking trade performance.

  • Owned product direction for an NLP trading engine trained on 40TB+ of LexisNexis global news data; drove integration of the AI core into the Tosan Galaxy automated trading service in partnership with the AI team.
  • Specified and co-designed a market-data warehouse ingesting feeds through RESTful APIs and Kafka for cleaning, validation and storage.
  • Delivered seasonal OKRs at 93% achievement rate; 50+ B2B customers; turnover in the tens of thousands of pounds.
Dec 2015 - Dec 2018
Tehran

Co-founder & Product Manager (Exited)

@ Ostadkar ↗

Market-leading home-services marketplace, connecting 1M+ customers with 50K+ local professionals.

  • Co-founded and built the product from scratch, running the market and customer research and competitor analysis, then validated the concept hands-on: prototyped and manually tested two service models in a controlled environment within weeks, pivoting to the one that worked. Designed the first UX and UI before any designer or engineer was hired.
  • Designed the professional ratings, badges and ranking-incentive system that drove supply quality (the core challenge in home services), a key factor in becoming the category leader.
  • Improved customer-app funnel conversion by 35% on average through 70+ A/B experiments and consistent data analysis.
  • Lifted the customer app to a 4.6 Google Play rating; oversaw three mobile apps, a customer web app and a back-office system.
  • Led the 20-person product, engineering and design team (in a 70-person company), owned the roadmap, and ran OKRs and KPIs across the business.
Sep 2012 - Nov 2015
Tehran

Machine Learning Engineer

@ Sharif Vision Tech

Computer vision company building commercial products across sports analytics, manufacturing and quality control.

  • Built the football player tracking platform end to end, from capture hardware and detection and tracking models to the reporting UI used by coaches and analysts. Sold through the Iran Football Federation and used to track around 300 matches across two operational years.
  • Delivered multiple other computer vision products including a banknote production quality control system, a tile sorting system and several other quality control deployments. Contributed across hardware design, the vision pipeline and the models in each one.
§ Education
2009 - 2013

Bachelor's Degree in Computer Software Engineering

@ University of Science and Culture, Tehran ↗

Designed and built a computer vision platform that detects, tracks and follows multiple people in real time, mapping their positions to a floor plan for indoor or outdoor spaces. Produced individual routes, dwell time at specific zones, entry and exit counts, and time-windowed statistics on footfall and location-to-location flow.

prefer the long version?
The CV has dates, references, and the full publication list.
Download CV ↓