Backend & Applied ML

Joshua Daniel Talahatu

Python-focused backend developer shaping retrieval systems, ML inference workflows, and portfolio demos that can be audited end to end.

See the work Resume Get in touch

About

Hello, I’m Joshua

I’m a backend and applied ML engineer focused on building practical systems that are as well engineered as they are intelligent.

I work primarily in Python, creating APIs, retrieval and search systems, reproducible evaluation workflows, and deployable demos that turn ML concepts into software people can actually use.

Stack and tools

PythonDjangoFlaskFastAPITensorflowPyTorchscikit-learnPostgreSQLLaravelDockerReactTypeScript

Surabaya, ID

Based in

Python

Focus

Portfolio-ready

Project posture

Experience

Work and Education

May 2024 - August 2024
Remote
Open Source Contributor - GSoC 2024
NumFOCUS / gems-uff noWorkflow
Extended Python provenance tooling with AST-level trial diffing and backward program slicing for reproducible computational workflows.
- Adapted Zhang-Shasha tree-edit distance to compare Python AST records stored in noWorkflow's SQLite provenance schema.
- Implemented backward slicing for the now show --slice workflow and improved query performance through targeted database indexes.
- Added deterministic synthetic-trial tests so AST comparison and slicing behavior could be verified without fragile fixtures.
2025 - 2026
Surabaya, Indonesia
Backend and ML Systems Builder
KerjaHebat AI recommender prototype
Built and audited a cold-start job matching pipeline that combines semantic retrieval, structured metadata, and baseline comparisons.
- Prototyped dense retrieval with multilingual E5 embeddings and FAISS candidate generation for resume and vacancy matching.
- Compared embedding retrieval against BM25, TF-IDF, rules, and traditional ML baselines to keep the portfolio claim evidence-backed.
- Built parsing and normalization endpoints for PDF/DOCX resumes with skill taxonomy mapping for downstream ranking experiments.
2025 - 2026
Surabaya, Indonesia
Predictive Maintenance Demo Builder
NIST Genesis anomaly detection service
Turned anomaly-detection notebooks into an API-first maintenance demo with explicit model configuration and ERP-style workflow hooks.
- Served a champion anomaly model through FastAPI using a runtime inference config instead of hardcoded notebook assumptions.
- Designed an idempotent ERP/CMMS webhook flow to create maintenance work orders without duplicate device-timestamp events.
- Documented a free-tier-friendly AWS Lightsail deployment path and Prometheus-style operational metrics for portfolio review.