╔══════════════════════════════════════════════════════════════╗
║ C O L A C B R R ║
║ · Digital alchemist / bug exorcist · ║
║ Python · FastAPI · AI/ML · Full-Stack · MLOps · GCP ║
╚══════════════════════════════════════════════════════════════╝
Backend-leaning software engineer building applied AI/RAG systems, full-stack prototypes, and integration-heavy products.
Python · FastAPI · React · Docker · RAG · MLOps · GCP
Hey. I'm Cristian-Adrian Colăcel, a software engineer from Bucharest currently finishing a Master's in Information Systems at POLITEHNICA Bucharest.
I build backend-heavy, product-shaped systems that work end-to-end: Python/FastAPI APIs, applied AI and retrieval pipelines, full-stack dashboards, and infrastructure that can actually run, be tested, and be debugged locally. My strongest direction is systems and integration engineering, with a growing focus on applied AI/ML engineering, retrieval systems, model serving, MLOps workflows, and cloud-native architecture.
- Built 4 public end-to-end systems spanning RAG, full-stack platforms, ML inference, and scientific benchmarking.
- Measured retrieval benchmarks: ~6.9 ms average search latency, Recall@10 0.88 on 1k images.
- Delivered production-adjacent experience in Django web development and cross-platform QA.
I am best matched with early-career software engineering roles that involve building practical systems end-to-end: APIs, product backends, dashboards, retrieval pipelines, model-serving workflows, automation, data flows, and service integration.
Strongest directions: backend engineering, applied AI/ML, systems integration, full-stack product engineering, data engineering, and cloud-adjacent platform work.
Web Development Intern · ImpexRegio | June – September 2024, Bucharest
Built Django backend features and SQL integrations for production website workflows. Delivered responsive frontend components connecting user-facing pages with backend functionality. Supported database design, query updates, debugging, and testing for full-stack web features. Managed GitHub branches and followed version-control workflows with the team.
Helped accelerate recurring website deliverables by approximately 5 days through coordinated task tracking, Git collaboration, and focused debugging support.
QA Game Tester · EA Games | June – December 2022, Bucharest
Dragon Age: The Veilguard (pre-alpha) · PC · PlayStation · Xbox
Executed functional, regression, and compatibility testing across three platforms during the production cycle. Identified, reproduced, documented, and tracked defects with clear reproduction steps and severity context in Jira. Supported build implementation, version tracking, and release workflows through Azure DevOps.
| Degree | Institution | Period | Focus |
|---|---|---|---|
| M.Sc. Information Systems | POLITEHNICA Bucharest | 2025 – present | Computer vision, image and signal processing, statistical modeling, simulation, scientific databases, multimedia systems, wireless networks, optimization, and nature-inspired computing |
| B.Eng. Computers and Information Technology | POLITEHNICA Bucharest | 2020 – 2025 | Software engineering, OOP, algorithms, data structures, AI fundamentals, pattern recognition, computational intelligence, Android development, database design, networking, distributed systems, parallel computing, embedded systems, HCI/UI design, cryptography, and data protection |
Backend / APIs
Python FastAPI Django Django REST Framework Flask Pydantic SQLAlchemy Alembic REST APIs WebSockets
Frontend / Product
React TypeScript JavaScript Next.js Vite SvelteKit Tailwind CSS Recharts D3.js Flutter / Dart
AI / ML / Data
PyTorch TensorFlow scikit-learn FAISS CLIP RAG MLflow Ollama Pandas NumPy SciPy OpenCV Numba
Databases / Storage
PostgreSQL SQL YugabyteDB SQLite Supabase MinIO Row Level Security
Infra / DevOps
Docker Docker Compose Nginx Linux Git GitHub systemd Bash
Cloud / GCP
Vertex AI Vertex AI Studio BigQuery Compute Engine APIs Explorer
Hardware / Embedded
ESP32 Raspberry Pi 5 IoT telemetry Sensor pipelines Servo motors Signal processing
Familiar / Academic
Go C/C++ C# Assembly R
What it is: Full-stack IoT and ML system for adaptive solar tracking, telemetry, forecasting, and dashboard monitoring.
Why it matters: It connects real hardware, embedded control, backend storage, live visualization, and ML forecasting into one working vertical slice.
Highlights
- Built dual-axis servo tracking with LDR arrays on ESP32, plus weather-protection logic for rain and vibration events.
- Streamed voltage, current, battery state, temperature, humidity, rain, and vibration telemetry to a Raspberry Pi 5 backend.
- Used Django, PostgreSQL, WebSockets, D3.js dashboards, and LSTM-based solar irradiance forecasting for next-day production estimation.
Stack: ESP32 Raspberry Pi 5 Django PostgreSQL D3.js WebSockets LSTM INA3221 DHT22 SG90 servos
What it is: Local-first semantic search over images and video using CLIP embeddings, FAISS indexes, FastAPI APIs, and a React interface.
Why it matters: It demonstrates a complete applied retrieval pipeline with multimodal indexing, reranking, explainability, and measurable retrieval quality.
Highlights
- Combined text, image, frame, and video retrieval with separate modality indexes and timestamped best-frame metadata.
- Added caption-aware reranking, Ollama-generated grounded explanations, prompt versioning, explanation caching, and Recall@K benchmarking.
- Measured ~6.9 ms average retrieval latency, Recall@10 of 0.88 on a 1k-image run, and Recall@10 of 0.57 on a 5k-image run.
Stack: FastAPI React PyTorch FAISS CLIP Ollama OpenCV
What it is: Full-stack e-startup and template marketplace with separate public, client, and staff workflows.
Why it matters: It models a realistic product backend with authentication, RBAC, transactional business flows, object storage, and operational automation.
Highlights
- Built cookie-based sessions, RBAC, login audit events, and optional 2FA model support across separate user flows.
- Wired checkout into a transactional sequence covering client profile, order, invoice, project, template assignment, site version, and template license.
- Used YugabyteDB for distributed SQL, MinIO for signed URL delivery, and Makefile automation for startup, reset, smoke testing, health checks, and reports.
Stack: FastAPI React TypeScript YugabyteDB MinIO Docker Compose Nginx SQLAlchemy Alembic
What it is: Applied ML platform for facial emotion recognition with training, comparison, and live inference workflows.
Why it matters: It turns model experimentation into a product-shaped system where multiple approaches can be compared through APIs and a dashboard.
Highlights
- Implemented classical baselines with HOG/LBP + SVM alongside deep learning models such as ResNet and EfficientNet.
- Exposed training, validation, model comparison, submission generation, and live inference through FastAPI endpoints.
- Built a React dashboard where model comparison is part of the product, not just a set of offline scripts.
Stack: FastAPI React PyTorch ResNet EfficientNet SVM OpenCV
What it is: Scientific Python CPU benchmark comparing vectorized NumPy and parallel Numba backends for a 2D FitzHugh-Nagumo simulation.
Why it matters: It shows practical Python performance engineering: measuring when vectorization is enough and when JIT parallelism is worth the warm-up cost.
Highlights
- Benchmarked both backends with identical equations, initial state, grid sizes, and iteration counts.
- Reported timing, ms/iteration, ns/cell/iteration, mean/max absolute error, speedup ratios, and paired t-test results.
- Packaged the workflow as a configurable CLI for reproducible simulation and benchmark runs.
Stack: NumPy Numba SciPy Matplotlib CLI
Product-oriented experiments and supporting projects. Actual usable software.
| Project | What it does | Stack |
|---|---|---|
| Mall Customer Segmentation | Python + R segmentation project with aligned PCA + K-Means workflows, report generation, and a FastAPI + React automation layer | Python · R · scikit-learn · FastAPI · React · nbconvert · RMarkdown |
| YouTube Transcript Pipeline | CLI pipeline for scraping, normalization, deduplication, dataset assembly, TF-IDF baseline, and BiLSTM training | Python · Playwright · scikit-learn · TensorFlow/Keras · LSTM · Pandas |
| InvestSim | Browser-based personal finance simulator for compounding, contributions, inflation-adjusted value, withdrawals, IRR, scenarios, charts, and CSV export | Next.js · TypeScript · Tailwind · Recharts |
| CV & Cover Letter Generator | Profile-driven CLI that generates ATS-optimized CV and cover letter packages from raw job offers and grounded profile facts | Python · LaTeX · pdflatex · Markdown parsing · CLI |
| Remote-Terminal | Phone-to-Linux remote access workflow using Tailscale, SSH, tmux, service-state restoration, and operator dashboards | Linux · Bash · Tailscale · SSH · tmux · systemd · Python |
| Audio Analysis Platform | Real-time audio capture, streaming, FFT analysis, filtering, and live dashboard across hardware and browser | ESP32 · FastAPI · WebSockets · React |
| MLflow Model Lifecycle | End-to-end MLflow workflow covering experiment tracking, model registry, aliases, local inference, and served prediction requests | Python · MLflow · scikit-learn · SQLite |
| WardrobeApp | Wardrobe management with authenticated collection CRUD, private image storage, signed URLs, and outfit scaffolding | Next.js · Supabase · RLS · TypeScript |
| LeafPad | Windows-first offline PDF reader with progress persistence, bookmarks, rotation, zoom, and touch-friendly navigation | Electron · React · PDF.js · TypeScript |
| VocabMaster | Vocabulary learning mobile app with flashcard review, retention tracking, and study feedback | Flutter · Dart |
| Simple Daily Quotes | Flutter app with backend-connected quote fetching, ETag caching, offline fallback, and daily notifications | Flutter · Dart · SharedPreferences |
| Admitere Academia de Politie | Romanian Police Academy admission-prep platform with study content, quizzes, PDF viewing, and gamification | FastAPI · SvelteKit |
Currently working through Google Cloud Skills Boost, focused on certification preparation and hands-on labs around data engineering, cloud infrastructure, BigQuery, Vertex AI, load balancing, and RAG workflows. I am using these labs to build evidence toward deployable cloud projects, not presenting them as production cloud experience.
Completed · 100% assessment scores
| Lab / Course | Status |
|---|---|
| A Tour of Google Cloud Hands-on Labs | ✓ |
| APIs Explorer: Qwik Start | ✓ |
| Get Started with Vertex AI Studio | ✓ |
| Generative AI with Vertex AI: Prompt Design | ✓ |
| Navigate BigQuery | ✓ |
| Introduction to Generative AI | ✓ |
| Introduction to Large Language Models | ✓ |
| Introduction to Responsible AI | ✓ |
| Introduction to Data Analytics in Google Cloud | ✓ |
In progress
| Course / Path | Status |
|---|---|
| Preparing for your Professional Data Engineer Journey | In progress |
| Preparing for your Associate Cloud Engineer Journey | In progress |
| Create Embeddings, Vector Search, and RAG with BigQuery | In progress |
| Implementing Cloud Load Balancing for Compute Engine | In progress |
| Prompt Design in Vertex AI | In progress |
| Data Management and Storage in the Cloud | In progress |
Targets: Google Cloud Professional Data Engineer · Google Cloud Associate Cloud Engineer
I prefer real projects, readable code, measured benchmarks, and boring infrastructure that works.
Commit small.
Document the ritual.
Never trust a green build you didn't actually read.
- LinkedIn: linkedin.com/in/cristiancolacel
- For role-specific CVs or project details: reach me on LinkedIn
- Location: Bucharest, Romania
- Open to: early-career / graduate / internship roles in backend, AI/ML, systems integration, full-stack, or data engineering