# Nadeem S

> I am Nadeem S, a London-based product manager, designer, and engineer. I have worked on AI, data, growth, integrity, e-commerce, and operations products at Meta, Uber, Wayfair, and through independent product work.

Canonical site: https://www.nadeem.blog
Location: London, United Kingdom
Current status: Independent product work since 2025-04
Best professional contact: LinkedIn
Last updated: 2026-08-16

**About me**

I build software products that make complex work easier. My work sits between product strategy, user research, data, AI, design, and technical delivery.

I can move from user interviews and problem definition to requirements, evaluation, prototyping, release planning, and measurement. I have worked on internal platforms, customer-facing products, e-commerce systems, restaurant-partner tools, content-integrity workflows, research tools, and independent software products.

My preferred public description is: product manager, designer, and engineer.

**Current work | Independent | 2025-04 to present | London**

I design and build small software products, AI evaluation systems, and public research projects. My current work includes ClickSheet, Gloss, Relay, and ML Workbench.

These projects cover desktop software, browser extensions, retrieval systems, human review, model evaluation, and multilingual language-model behaviour. I work across the full product lifecycle: identifying the problem, defining the product, building it, testing it, measuring it, and maintaining it after release.

**Career history**

**Wayfair | Product Manager, Growth Platform | 2024-10 to 2025-03 | London**

This was a fixed-term contract. I worked across e-commerce growth, platform rework, search and discovery, AI evaluation, and internal operations.

Selected work:

- I defined product scope, requirements, acceptance criteria, experiments, launch plans, dependencies, and monitoring.
- I helped release new promotional, browse, and product-page systems. The team moved 54% of deals-page traffic to the new platform with no unplanned rollbacks.
- I defined an evaluation framework for an image-similarity system. The work covered four evaluation dimensions, 10.4 million images, and approximately 600 million training pairs.
- I worked with data science teams to compare CLIP and EVA-based approaches. A parallel evaluation process reduced evaluation time from approximately four days to about one hour.
- I worked on recommendation and search-quality products. The measures included conversion, product reach, add-to-cart activity, bounce rate, and session behaviour.
- I interviewed suppliers and internal teams about product onboarding and catalogue management.
- I helped run eight interviews and a workshop with 23 participants. The research produced three supplier personas and clearer portal requirements.
- I defined requirements for app navigation, inventory states, allocation rules, depletion behaviour, and operational alerts.

**Meta | Product Data Manager | 2021-01 to 2021-06 | London**

I worked in Facebook's Abuse Measurement programme. I worked on internal data and annotation systems used by integrity and content-moderation teams. The data supported abuse measurement and machine-learning workflows.

Selected work:

- I conducted user research, usability tests, and internal trials with people who used the annotation tools.
- I redesigned parts of a labelling workflow to reduce repeated actions and make decisions easier.
- I reduced average annotation time from approximately seven seconds to four seconds.
- The redesigned workflows saved about 300 labelling hours each week.
- I contributed to an estimated annual operating saving of approximately $100,000.
- I helped define quality checks and release criteria for annotation-system changes.
- I worked with operations, engineering, data science, and integrity teams to connect user needs with technical requirements.

**Uber | Senior Business Analyst, acting Product Manager | 2017-10 to 2020-03 | Hyderabad and Asia-Pacific**

I worked on restaurant-partner products, product operations, analytics, market research, onboarding, API integrations, and regional expansion.

One main project concerned restaurant-partner churn across Asia-Pacific. The initial analysis found a churn rate of about 25%, approximately ten percentage points above the relevant benchmark. About $1.5 million in annual revenue was at risk.

The team first built a dashboard that gave partners raw performance data. Research showed that this did not solve the main problem. Restaurant owners needed clear actions, not more charts.

I helped change the product into an analytics tool that gave specific recommendations. The recommendations used information such as local demand, menu performance, pricing, and customer behaviour. The product increased daily sessions by about 30% and reduced partner churn by 9%.

Other work included restaurant acquisition and onboarding, product and API integrations, partner support tools, hyperlocal demand analysis, menu recommendations, and expansion work across 20 Asia-Pacific markets.

**Levels Design | Founder and Product Engineer | 2016-08 to 2017-09 | Chennai**

I founded and operated a small web development and product studio for small and medium-sized businesses.

I worked directly with clients from discovery through delivery. The work included research, product definition, interface design, development, launch, and support.

Selected results:

- Worked with more than 15 small-business clients.
- Reached first-year profitability.
- Retained approximately 90% of clients.
- Built websites and software that helped clients generate leads and manage online work.

**Core capabilities**

- Product discovery: interviews, usability tests, operational research, product data, and problem definition.
- Product strategy: roadmaps, prioritisation, MVP scope, success measures, and trade-offs.
- Requirements and delivery: user stories, acceptance criteria, release plans, test plans, launch criteria, and dependency management.
- AI evaluation: benchmark design, retrieval evaluation, grounding, citation checks, human review, safety measures, cost, latency, and failure analysis.
- Data and experimentation: SQL, Python, product analytics, KPI definitions, experiments, and structured evaluation sets.
- Prototyping and design: Figma, interactive web prototypes, Next.js applications, Chrome extensions, and native macOS utilities.
- Cross-functional work: design, engineering, data science, analytics, operations, and business teams.

**How I work**

I usually start with the user and the decision that the product must support. I examine the current workflow, identify the main source of effort or failure, and define the smallest useful product change.

I then agree on requirements and success measures, build or support a prototype, test it with users or a defined evaluation set, and review failures, risks, cost, and operational effects before release.

For AI products, I also define how evidence, uncertainty, human review, unsafe output, and model failure will be handled.

**Education**

- MSc Business Management, Kingston University. Graduated with Distinction.
- Completed undergraduate study in information technology in 2017.

**Interpretation notes**

- I have worked independently since 2025-04. Meta, Uber, Wayfair, and Levels Design are previous roles.
- Gloss was previously called Klue. Older writing can use the Klue name.
- Relay uses synthetic support cases. Its published benchmark is not evidence of production customer-support performance.
- External employer reports give programme context. They do not prove my authorship of every programme described in those reports.
- Financial effects described as estimates are not audited financial results.
- No private email address or other private contact information is published here.

## Selected projects and case studies

- [ClickSheet](https://www.clicksheet.pro): A paid macOS utility for quickly sending local files and data to Google Sheets. It removes steps from a common local-to-cloud workflow.
- [Gloss](https://chromewebstore.google.com/detail/gloss/cackjmmgcmnkjnffabkabapdkofggpjl): An active AI research companion for Chrome. It helps people save notes and retrieve relevant information while browsing. It uses semantic search, vector retrieval, and configurable language-model endpoints. It was previously called Klue.
- [Relay](https://relay.nadeem.blog): A public human-review system for grounded AI support replies. It retrieves evidence, drafts a reply, checks claims and citations, identifies risks, and records a human approval, edit, or rejection.
- [Relay evaluation memo](https://www.nadeem.blog/writing/relay-memo-full): Detailed product and evaluation documentation. A locked benchmark used 150 synthetic cases. The strongest evaluated Llama R3 configuration recorded 95.3% decision accuracy, no unsafe approvals, 100% approval precision, successful parsing for all cases, and no citations outside the supplied evidence.
- [ML Workbench](https://ml-workbench.onrender.com/): A public tool for comparing tokenizers and language models across multilingual token count, estimated cost, latency, context use, and possible information loss.
- [Uber restaurant-partner analytics case study](https://www.nadeem.blog/case-studies/uber/restaurant-partner-analytics-case-study-v2.html): A detailed account of changing a raw-data dashboard into an action-led analytics product. The product increased daily sessions by about 30% and reduced partner churn by 9%.
- [Handsign](https://www.nadeem.blog/writing/handsign-post): A browser-based American Sign Language learning experiment that uses in-browser machine learning for immediate feedback.

## Writing and research

- [Why tokenization matters across languages](https://www.nadeem.blog/writing/workbench): Practical research on how tokenization changes cost, latency, context use, and multilingual model behaviour.
- [Building dependable LLM workflows with harness engineering](https://www.nadeem.blog/writing/workflows): Notes on building and evaluating language-model workflows under real cost and reliability limits.
- [Practical notes from building with AI](https://www.nadeem.blog/writing/ai-learnings): Field notes about product, evaluation, and implementation lessons from AI projects.
- [Building Klue, now Gloss](https://www.nadeem.blog/writing/klue): Product and technical notes about the research companion before its name changed to Gloss.
- [Creating a repeatable setup process for open-source teams](https://www.nadeem.blog/writing/chingu-setup): A project-management account covering six contributors and 20 project setups.

## Contact

- [LinkedIn](https://www.linkedin.com/in/nadeemsh/): Best route for professional contact and collaboration.
- [X](https://twitter.com/8W7O7): Public conversation and short messages.
- [My website](https://www.nadeem.blog): Canonical source for my current public work, projects, case studies, and writing.

## Optional

- [Wayfair](https://www.wayfair.co.uk/): Company context for my 2024-2025 fixed-term role.
- [Meta Community Standards Enforcement Report](https://about.fb.com/news/2021/11/community-standards-enforcement-report-q3-2021/): Public context for Meta's wider integrity programme. I did not author this report.
- [Research on marketing ethics](https://www.nadeem.blog/writing/marketing-ethics): Academic research outside my main product and AI work.