Taner Behmen

About

Chemistry and Mathematics student who builds production software for a living.

I currently work as an AI Engineer at Gradient House, a Warsaw-based AI consultancy, where I'm the sole engineer running production systems end to end for enterprise clients — from the first architecture decision to the client demo.

My day-to-day involves building AI pipelines with verification built in, shipping software that has to be right, not just plausible, before it reaches a client. Before that, I was the sole technical operator at an early-stage startup, building its first production application and an automation stack from scratch.

Outside of work, I care about applying quantitative reasoning to real problems. My background in Chemistry and Mathematics gives me a systematic approach to analysis, whether that's in data, in code, or in a room with a team trying to figure something out.

Location

St Andrews, Scotland

Current role

AI Engineer, Gradient House

Studying

Chemistry & Mathematics, St Andrews

Open to opportunities

Projects

Agentic Data Analysis Platform

An agentic pipeline that turns raw client datasets into finished analysis — planning, running, and verifying each step before it's delivered, with automated report export. Built at Gradient House and refactored to generalise across datasets rather than being hardcoded to one client.

AI Agents Python Data Analysis Confidential

Enterprise Catalog Search

A search pipeline built for an enterprise client's product catalog of 8,800+ items, combining multiple signals to handle messy, typo-heavy queries far more reliably than any single method alone. Built at Gradient House.

Search Python Confidential

VibeTrader

An autonomous multi-agent trading system built for a private client. Several AI analysts debate the case for and against a position before a portfolio-manager agent issues a final call, with hard risk limits gating every trade.

AI Agents Python Trading

Private Invoicing Contract

An agentic research and invoicing pipeline built under a private contract — automatically researching and generating client invoices end to end.

AI Agents Automation Confidential

Zombie Delete

Cryptographic proof-of-deletion standard for enterprise and blockchain, giving independent proof of erasure across 20+ chains and cloud storage. Demoed at Web Summit Qatar, Wolves Summit semi-finals, and The Pitch Berlin.

ICP / Blockchain React.js Compliance

Bonded — POHR Protocol

Proof of Human Relationship protocol enabling verifiable, privacy-preserving evidence of personal relationships for immigration, insurance, healthcare, and legal use cases. 2nd place at the ICP Global Hackathon UK qualifier.

Web3 Identity LegalTech

Bonded Soulbound Token

A two-person, non-transferable relationship token built on the Internet Computer Protocol, designed and built directly with the Together Alone Ventures founder as a proof-of-relationship primitive.

ICP / Blockchain Web3

Multi-agent AI stack

An autonomous multi-agent system automating finance processing, investor relations, content production, and CRM operations end to end.

AI Agents Python Automation Confidential

Sales tracking tool

Custom sales analytics system for a Warsaw local business. Analysed transaction patterns to optimise staffing, reduce food waste, and increase revenue. Surplus profits directed to Turkish earthquake relief.

Python Data analysis

Experience

Current Jun 2026 – Present
GradientHouse Poland

AI Engineer

Warsaw, Poland · Remote

Sole engineer running two concurrent production systems for enterprise clients at a Warsaw AI consultancy. Own each system end to end, from first architecture decision to client demo.

Architected an LLM tool-calling pipeline that separates reasoning from computation — queries route across 26 typed analysis functions executed deterministically in Python, with an output-verification layer that rejects any figure not traceable to a tool result, reaching ~95% end-to-end accuracy on a 32-case eval suite
Refactored a single-client prototype into a dataset-agnostic platform using runtime data profiling and declarative config, cutting new-account onboarding down to one config file backed by 120+ automated tests
Built a hybrid lexical-semantic retrieval system for an 8,800-item product catalog with confidence-gated score fusion and a fine-tuned multilingual embedding model
Designed a rule-based chart recommendation engine to replace ad hoc LLM chart selection, and instrumented a production order-processing service with OpenTelemetry/Azure App Insights tracing plus a non-prod dry-run mode

AI Agents Python LLM Pipelines Search Automation

Jan 2026 – Jun 2026
Together Alone Ventures

Developer / Sole Technical Operator

Internship · Remote

Sole technical operator at an early-stage privacy-tech startup on the Internet Computer Protocol. Built the company's first production application end to end, then designed and ran the autonomous back-office infrastructure behind it.

Built the React.js frontend for Zombie Delete, a cryptographic proof-of-deletion tool, integrating it with ICP smart contracts to generate Cryptographically Verifiable Deletion Receipts (CVDRs); delivered live demos at Web Summit Qatar, Wolves Summit semi-finals, and The Pitch Berlin
Deployed a 5-agent orchestration stack on containerized decentralized cloud (Akash) covering outreach, content generation, CRM, and operational workflows, with sandboxed workspaces, scoped tool access, and scheduled execution
Diagnosed a silent failure mode where a production system reported false success on schema mismatch; rebuilt ground-truth verification into every automated workflow
Built data pipelines and dashboards for investor demos, and automated invoice processing for two legal entities (161 invoices: OCR, dedup, entity routing)

AI Agents React.js ICP / Blockchain Cloud Automation

Jul 2025 – Jan 2026
Bonded

Grant & Investment Researcher

Part-time · St Andrews

Research and investor outreach for a Web3 identity startup building a Proof of Human Relationship protocol. Focused on market sizing, competitive positioning, and identifying capital sources suited to a privacy-preserving identity product.

Mapped 40+ investors and grant programs aligned to a Web3 identity startup and built a structured outreach pipeline
Used Python and Pandas to quantify market size in identity verification and benchmark direct competitors
Produced pitch deck materials covering market sizing, competitive landscape, and strategic positioning
Contributed research that supported Bonded's 2nd place finish at the ICP Global Hackathon UK qualifier

Python Pandas Market Research Web3 Identity

2021 – 2023
Warsaw

Founder — Help4Refugees

Non-profit · American School of Warsaw

Founded and ran a humanitarian initiative to collect and ship computers to Ukrainian families displaced by the war. Started as a school project at 16; grew into a city-wide operation with corporate partners and two years of sustained fundraising.

Secured hardware donations from Warsaw-based companies and schools, shipping 100+ devices to displaced families
Organised multi-year fundraising events across Warsaw, growing the project beyond the school community
Built partnerships with local organisations to coordinate logistics and distribution
Directed surplus proceeds to Turkish earthquake relief efforts in 2023

Non-profit Fundraising Community Partnerships