Yaoundé, Cameroon · Open to remote / relocation
Noa Frédéric
Backend systems and applied NLP. I build the layer between raw data and structured output. Symptoms become diagnoses, English becomes Igbo, wallets move across borders. Every system ships with measurable outcomes, not vibes.
1for t in range(len(target)):2 scores = attention(decoder_state, encoder_outputs)3 weights = softmax(scores)4 context = weighted_sum(weights, encoder_outputs)5 decoder_state = decode_step(context, decoder_state)6 output[t] = predict(decoder_state)- API latency cut
- 42%
- answer accuracy
- 93%
- En→Igbo, vs NLLB-200 & OPUS-MT
- BLEU 40.67
- throughput
- 3×
Extracting signal
from noise.
Backend engineer and applied NLP researcher. I take systems end to end: REST APIs, data pipelines, model training, mobile builds, and cloud deployment on AWS EC2 and DigitalOcean.
My work sits where clean signal has to be extracted from noisy input: free-text symptoms mapped to candidate diagnoses, English mapped to severely under-resourced Central African languages, and cross-border transactions routed through compliance logic. I care about measurable outcomes: a 42% latency cut, 93% answer accuracy, BLEU 40.67 on En→Igbo, because precision is the actual deliverable.
Off the keyboard: Algorithms as a discipline (Stanford Online, 2024) and the patience that judo demands, both reward staying structural under pressure.
Four roles, one through-line.
Backend Engineer (Lead Contributor)
AFreeServ
Sep 2025 – Nov 2025RemoteReal-time messaging backend for a service-booking platform serving thousands of concurrent users.
- Built the real-time messaging system powering live conversations on a service-booking platform serving thousands of concurrent users.
- Cut API latency by 42% by rewriting slow queries, adding TypeORM indexes, and introducing a caching layer.
- Brought test coverage to ~85% with Jest and Supertest ahead of a major production release.
- Refactored core backend modules and raised overall throughput 3×; ran code reviews to keep the codebase consistent.
42%latency cut~85%test coverage3×throughputNestJSTypeORMRedisJestSupertestBackend Engineer
Ukuqala Platform · Clinical Triage System
Oct 2024 – PresentRemotecurrentClinical triage backend that analyses patient symptoms and surfaces structured guidance — the physician makes every final call.
- Designed and built the backend for a clinical triage platform that analyses patient symptoms and surfaces structured guidance to clinicians, with the physician making all final calls.
- Wrote the REST API layer in Python handling real-time intake of symptom data and returning structured clinical suggestions.
- Trained and integrated NLP models to parse free-text symptom descriptions and map them to candidate diagnoses.
- Built the full data pipeline from raw input through preprocessing, normalisation, and semantic similarity scoring.
- Built the mobile app frontend in Expo (React Native), covering patient symptom intake and the clinician dashboard.
- Set up authentication and data-handling controls aligned with healthcare privacy requirements.
- Loaded 583+ medical reference documents into Pinecone and tuned retrieval to keep responses under 2 seconds under load.
93%answer accuracy583+medical docs indexed<2sresponse under loadPythonFastAPIPineconeExpoReact NativePython Backend Engineer
Primus Cloud Solutions Ltd · Sendi — international money transfer
Aug 2024 – May 2025RemoteCore transaction backend and cross-border compliance for Sendi, an international money-transfer platform.
- Worked on Sendi, an international money-transfer platform. Built the core transaction backend and cross-border compliance logic.
- Wrote the Stripe payment microservice from scratch, adding fraud-detection rules that brought failed transactions down by 28%.
- Built the notifications service (email, SMS, push) with delivery latency consistently under 1 second.
- Set up a horizontally scalable microservice stack using FastAPI, Redis, and PostgreSQL.
- Deployed and maintained services on AWS EC2 and DigitalOcean — networking, environment configuration, and uptime.
−28%failed transactions<1snotification latencyFastAPIRedisPostgreSQLStripeAWS EC2Research Assistant, NLP & Data Science
University of Hagen & ICT University
Jul 2024 – Feb 2025RemoteLinguoMT — neural machine translation for low-resource African languages. Published in Informatics (MDPI).
- Built LinguoMT, a neural machine translation pipeline fine-tuned on MarianMT Transformer architectures across 9 African language pairs — Igbo, Hausa, Swahili, and severely under-resourced Central African languages such as Ghomala, Bulu, and Fulfulde Adamawa.
- Designed an automated data curation and alignment pipeline tailored to low-resource languages; training sets were capped at 10,000 sentence pairs per language for controlled, comparable experiments.
- Achieved BLEU 40.67 on English-to-Igbo, outperforming both NLLB-200 and OPUS-MT baselines, and 22.43 on English-to-Bulu.
- Established the first publicly available BLEU benchmarks for several Central African languages.
- Released all curated datasets, tokenizers, and fine-tuned models on Hugging Face Hub, and shipped a web-based real-time translation app alongside the research.
40.67BLEU En→Igbo9language pairs22.43BLEU En→BuluMarianMTTransformersTensorFlowHugging Face
Five systems,
measured outcomes.
LinguoMT
Neural machine translation for nine low-resource African language pairs, with first-of-kind BLEU benchmarks for Central African languages.
Ukuqala Clinical Triage
Clinical decision-support backend that turns free-text symptoms into structured guidance for physicians — who keep the final call.
Sendi
International money-transfer platform: core transaction backend, cross-border compliance, and a fraud-aware Stripe microservice.
AFreeServ Messaging
Real-time messaging system for a service-booking platform serving thousands of concurrent users.
AlgoViz
FAANG interview-prep platform covering 25 DSA problems with step-through algorithm animations and multi-language code examples.
First-of-kind BLEU
for Central African NLP.
Neural Machine Translation for Low-Resource African Languages (LinguoMT)
Establishes first-of-kind BLEU benchmarks for several Central African languages using a MarianMT-based pipeline with an automated low-resource data curation process.
- 2024AlgorithmsStanford Online
- Python for Everybody & Machine LearningGoogle / Coursera
- 2024TOEFL iBT — English ProficiencyETS
The actual toolkit.
Languages
05Frontend & Mobile
03Backend Frameworks
04AI & NLP
03Databases
03Cloud & DevOps
04Contact
Have a signal worth
decoding?
Backend systems, applied NLP, clinical AI. The fastest path is email, or leave a message and I’ll reply.