Personal details
Martin T.
Technical skills
About
Senior Software Engineer with 8+ years designing, shipping and operating production backend and data systems in Python (FastAPI, Django) on AWS/GCP. Owns services
end to end — API and data-model design, testing, CI/CD, observability, cost and latency budgets. Since 2022, applies LLM and retrieval technology as one component
inside well-engineered systems, treated with the same discipline as any other dependency: contracts, evaluation harnesses, regression tests, fallbacks and cost controls.
Fluent in English and French, remote-first across EU/US/APAC time zones.
Engineering practice: trunk-based development, typed Python, mandatory code review, test pyramid (pytest, contract and integration tests), ADRs for architecture
decisions, structured logging and tracing. Uses AI coding assistants (Claude Code, Copilot) as measured accelerators behind review and test gates — never as a
substitute for design or verification.
Work Experience
Document-grounded conversational assistant (RAG) — legal & regulatory domain
Replaced manual multi-document search with a single citation-grounded answer, by building a Q&A service (FastAPI, PostgreSQL, Qdrant, Redis) combining semantic
chunking, hybrid BM25 + dense retrieval and answer-level source attribution.
Cut LLM inference spend 55% while removing single-vendor dependency, by designing a provider-agnostic gateway (OpenAI, Anthropic, DeepSeek, self-hosted
Ollama/vLLM) with cost-aware routing, automatic failover and per-tenant rate limiting behind one internal contract.
Kept the service running unattended in production, by containerising the stack (Docker Compose, Traefik, TLS automation) and adding GitHub Actions CI/CD with
migration gating, automated backups and periodic restore drills.
Prevented regressions across model and prompt upgrades, by building an evaluation harness (LangSmith, golden-question set, RAGAS-style scoring) run in CI on every
release candidate.
Market-intelligence & product-research platform for e-commerce merchants (Viseon.ai)
Kept daily ingestion reliable across multiple marketplaces despite hostile scraping conditions, by productionising distributed Playwright/Scrapy crawlers on Airflow with
retry/backoff, proxy rotation, anti-bot handling, schema validation and alerting on data-freshness SLAs.
Rebuilt the platform (Django/DRF + Next.js) for Shopify merchants, by redesigning the data model and query layer for catalogue-scale filtering and adding OpenAI-
backed generation of product copy and ad creatives.
Reduced deployment time 60% and eliminated manual release steps, by moving to Docker images with GitHub Actions CI/CD and environment-parity configuration.
over OpenAI APIs and LangChain.
Built a multi-tenant conversational platform for business workflow automation, by designing the Django backend, conversation-state persistence and tool-calling layer
Kept production availability high across releases, by owning AWS infrastructure and GitHub Actions pipelines with staged rollouts, health checks and rollback
procedures.
Improved intent-routing reliability, by restructuring the NLP classification layer and adding a labelled regression set to the test suite.