David Rengifo
AI Engineer · LLMs · RAG · Embeddings · Data Pipelines
AI engineer building end-to-end intelligent systems — from data pipelines to production LLM applications. Scientific research background (immunology, oncology, virology) adds a hypothesis-driven approach to model design and validation. Specializing in RAG architectures, vector search and embedding AI into robust data pipelines.
AI & ML
Data & Backend
DevOps & Cloud
Visualisation
Professional Experience
IoT R&D Engineer — Work-Study
2023–2025
SCLE SFE · Marseille — Railway IoT
Embedded C · LoRaWAN · Docker · Safety-critical
- Built and integrated a C error-handling library (STM32) for an axle-counting IoT system over LoRaWAN — achieving <3s end-to-end latency in production across multiple railway sites
- Designed and executed 100+ software test cases with full traceability documentation (test plans, specs, reports)
- Managed deployment pipeline (Docker, SSL) for PHP monitoring interface across multiple live production sites
Research Engineer — Oncology, Virology & Data Analysis
2019–2022
CIML Marseille · APHM/IHU · Harvard Medical School CBDM (Boston)
High-dim data · Clustering algorithms · Statistical modelling · COVID-19 diagnostics
- Developed a custom clustering algorithm in R for multi-parameter cytometry data (FACS + ZeNITH) — unsupervised ML on high-dimensional biological datasets applied to brain cancer characterisation
- COVID-19 diagnostics, variant detection/reporting and kit evaluation at APHM/IHU during the pandemic — rigorous data traceability and procedure writing for ARS compliance
- Research internship at Harvard Medical School (CBDM, Boston) — FACS analysis of RORγ+ Treg lymphocytes in tissue repair models
Projects — 2023–2026
⭐ Production · RAG Pipeline · LiveCoinsProphet — AI Crypto Intelligence Platform coinsprophet.davrengifo.dev
Python · FastAPI · pgvector (HNSW) · OpenAI Embeddings · gpt-5.4-mini · Redis · Docker Compose · GCP · GitLab CI/CD
End-to-end RAG pipeline: real-time crypto news (3 RSS feeds, 15-min) + price data (CoinGecko, 5-min, 8 tokens) → OpenAI text-embedding-3-small (1536-dim vectors) → pgvector HNSW cosine index → similarity search → gpt-5.4-mini response with cited sources. Daily AI market report at 08:00 UTC (dedicated report_generator service). FastAPI /ask RAG endpoint + /report endpoints. React 18 ChatRAG UI with source citation chips. 10 microservices via Docker Compose on GCP e2-micro VM, Nginx HTTPS reverse proxy (Let's Encrypt), GitLab CI/CD.
GPT-4o-mini · FastAPI · React 19 · TypeScript · Pydantic v2 · Railway · Vercel
Production-grade AI copilot with 3 NLP tools: customer feedback analysis (sentiment, themes, priority KPIs), document summarization (PDF support, up to 100k chars), and entity extraction (dates, names, amounts). Multi-step prompt engineering with structured JSON output validated by Pydantic schemas. API key auth, rate limiting (slowapi), OpenAI spend cap. Full test suite: 46 tests (20 pytest + 26 Vitest/RTL). Deployed on Railway + Vercel with custom domain.
LoRa · C embedded · Python · Backend · Oracle Cloud (OCI) · Web/Mobile dashboard
IoT lead within a 10-person team (IoT sub-team of 3→4). Designed full LoRa sensor network, backend data pipeline and real-time colony monitoring dashboard. Functional MVP delivered — Epitech graduation project winner.
AI — Image Classification (Pneumonia detection)
Python · CNN · Scikit-learn · PCA · Decision Tree
ML models for X-ray pneumonia detection. Benchmarked Decision Tree, PCA and CNN approaches with reproducible evaluation pipeline and comparative performance analysis.
AI — NLP & Optimal Routing
Python · Flask · Flair · Levenshtein distance · Dijkstra
Voice-command Flask app (French speech-to-text). Entity matching via NLP + Levenshtein distance. Optimal route computation using Dijkstra algorithm.
AI — Reinforcement Learning
Python · SARSA · Q-learning · Taxi-v3
Implemented SARSA and Q-learning agents on Taxi-v3 environment. Hyperparameter optimisation (α, γ, ε) with performance benchmarking and convergence analysis.