Hire the Best Digital Signal Processing Specialists

As a solution architect I have 14+ years of ASIC/FPGA/SOC based Design/Develop/Deployment experience from user requirement to the end system. -- Digital Design of complex blocks, RTL design, systemVerilog/IPcore Design -- FPGA emulation of large ASIC designs on Dense Xilinx FPGA chips -- Xilinx SOC based digital designs using Vivado design suit. -- High speed ethernet/PCIe Designs ADC/DAC Software definer radios DSP cores -- Embedded Linux - devicetree - driver - usersapce apps -- Timing closures - CDC based designs -- High speed ADC/DACs from ADI designs -- Verification using systemVerilog and cocoTB Tools: -- Xilinx FPGA Vivado Design Suite, Debugging using ILA Xilinx ZYNQ SOC / Ethernet / DMA / cores. -- IPs from Cadence, precise-ITC, and ARM, as well as VIPs from Mentor Graphics, Synopsys, and TrueChip. With a high level of skill, responsibility and dedication to solution delivery, I always realize every requirement of the client to the maximum in detail, and guarantees the timeline of the project with excellent time management skills.

I am a seasoned programmer and AI specialist with over 6 years of experience, backed by a professional team. My expertise centers on developing advanced solutions in Artificial Intelligence, Machine Learning, Deep Learning, MATLAB, and Python. I excel in designing and implementing complex algorithms, data analysis, and model optimization to solve real-world problems. In addition to my technical skills, I offer mathematics tutoring and problem-solving support to help you grasp core concepts and enhance your project outcomes. If you're looking for expert guidance or collaboration in AI, ML, MATLAB, or Python development, feel free to contact me. Committed to delivering high-quality, innovative, and professional results.

I'm an Upwork member with 100% Job Success and Top Rated Plus Python and C++ programmer with extensive experience in sound, image and video processing. I have been working for a long time with the OpenCV, GStreamer and NVIDIA Deepstream SDK libraries, using GPU hardware with CUDA and other HPC libraries. My work experience includes professional video players & editors, transcoders (using different types of inputs, such Blackmagic Decklink capture cards and all the video formats: HLS, SRT, fMP4, WebRTC), DSP systems, JUCE framework, object detection and tracking, Deep Learning systems (YOLO, Pytorch, Tensorflow, TensorRT), AI hardware like NVIDIA Jetson Orin and Raspberry Pi devices. I have also experience in ultrasound signal processing. In ML and LLM models I'm working training Reinforcing Learning PPO models and LORA fine tuning methods on AI. Also I have experience with TTS and Speech Recognition systems using models like Whisper, WhisperX and PiperTTS. In Hardware and PCB design I have good experience in Arduino and IDF Espressif systems, based on ESP32, ESP32s2 and ESP32s3 chips with standard thermal cameras and ESP32 Drones, using MQTT messaging and SSL certificates (WiFi, Bluetooth and external sensors). I can design and build a custom PCB board. I like to choose projects that interest me and that match my skills and give me a challenge or bring something new.

Machine learning engineer for time series and sensor data. I build and validate predictive pipelines on biosignals and wearables, IoT streams, energy series, and time-indexed text. Every engagement runs end to end, from raw data through validated model to a report and codebase your team can run without me. WHAT THE WORK LOOKS LIKE Signal processing and feature extraction from noisy multichannel data. Forecasting, classification, and anomaly detection under time-aware and participant-level cross-validation. Data pipelines that turn raw sensor streams or document archives into analysis-ready, versioned datasets. Statistical testing and signal analysis when the question is "is this effect real" rather than "build me a model." Deep learning where it earns its place and classical methods where they don't need replacing. Results reported with uncertainty, not just a headline metric, and delivered as reproducible, containerized Python code, deployed where the engagement calls for it, with technical reports written for both engineers and decision-makers. The center of my practice is time series and sensor data. I've applied it across biotech and wearables, finance, energy, and industrial IoT, and the validation discipline is the same in every one: the wrong split quietly invalidates the result, whichever industry it's in. SELECTED OUTCOMES ◆ Improved heart-failure readmission prediction by over 20% AUC from chest-worn accelerometer data, validated with patient-level cross-validation and confidence intervals, so the Series A medtech team had a result they could defend to clinical and investor audiences. ◆ Turned 18 years of editorial coverage into a quantitative geopolitical index for institutional finance, with an entity-to-equity layer carrying a point-in-time firewall (zero look-ahead events across 46,000 article-entity rows) and lead-lag relationships validated in-sample against country ETFs. Licensed to an institutional quantitative fund. ◆ Built a production-ready IoT demand-forecasting system: 10–13% MAPE under walk-forward validation and a consumption-event detector at 93.2% accuracy, handed off with the PoC-to-MVP deployment roadmap. ◆ Benchmarked seven deep-learning forecasters against classical baselines for solar irradiance under nested time-series cross-validation, with Diebold-Mariano tests and multiple-comparison correction. The finding the client could act on: deep models earn their cost at 1-hour and multi-day horizons, but at day-ahead a seasonal baseline is within noise, so the recommendation was a cheaper model for that horizon. Publication-grade report with full statistical appendix. ◆ Audited an inherited deep time-warping model and found it was optimizing the wrong objective; re-implemented it with an attention-based architecture, multi-seed evaluation, and confidence intervals on retrieval metrics, turning an unverifiable prototype into a reportable one. ◆ Designed and ran a falsifiable research pilot on sleep EEG testing whether model-error divergence precedes classical early-warning signals, with residual diagnostics and sensitivity analysis; the client called it "exceptional professionalism, rigor, and methodological clarity." HOW ENGAGEMENTS RUN Fixed-price, scoped milestones. Larger projects start with a paid discovery phase that produces a technical plan, data assessment, and acceptance criteria before any modeling begins. Shorter formats for well-defined questions: one-to-three-day signal or statistical analyses, technical reviews of an existing pipeline or model, and consultations on data strategy, feasibility, and what to build first. TOOLS Python (NumPy, SciPy, pandas, scikit-learn, statsmodels, PyTorch, LightGBM, XGBoost, Optuna, RDKit), signal processing (filtering, EMD decomposition, spectral and time-frequency features), pvlib, Hugging Face Transformers, SQL, Docker, FastAPI, AWS Tell me what you're trying to predict, understand, or decide, and what data you have or plan to collect. I'll identify the most practical first step, and if it's a fit, I'll scope it clearly.