[ { "title": "Quiver", "featured": true, "category": "Open Source", "range": "2024 – Present", "role": "Author", "summary": "Security-first vector database written from scratch in Rust. ANN search (HNSW, DiskANN, IVF + quantization), SIMD distance kernels, and end-to-end encryption built into the core.", "highlights": [ "32× less RAM at equal recall via DiskANN + PQ (10M × 768-d: ~1 GB vs ~31 GB).", "Ranked 2nd on SIFT1M behind FAISS, ahead of Qdrant, Milvus, Chroma, pgvector, LanceDB, and Weaviate.", "XChaCha20 encryption, RBAC, and immutable audit log.", "Python SDK, gRPC + REST API, ratatui TUI cockpit, and MCP server." ], "stack": ["Rust", "HNSW", "DiskANN", "IVF", "SIMD", "XChaCha20", "gRPC", "Python SDK", "MCP"], "impact": "32× RAM reduction at equal recall. 2nd on SIFT1M.", "link": "https://github.com/achref-soua/quiver", "linkLabel": "View on GitHub" }, { "title": "Helio", "featured": true, "category": "Open Source", "range": "2025 – Present", "role": "Author", "summary": "Self-hostable, AI-native marketing platform — open-source alternative to HubSpot and Klaviyo. Polyglot stack (TypeScript + Python), durable journey execution, and an AI copilot that turns plain language into segments and campaigns.", "highlights": [ "Multi-tenant (Postgres RLS), event pipeline Redpanda → ClickHouse, journeys on Temporal (durable, crash-safe).", "AI copilot: build segments and journeys in plain language. Churn scoring via ONNX + XGBoost.", "MCP server — external AI agents can launch campaigns programmatically.", "229 PRs, 120 E2E tests, 9 production container images, zero Trivy critical findings." ], "stack": ["TypeScript", "Next.js", "Python", "FastAPI", "Temporal", "ClickHouse", "PostgreSQL", "Redpanda", "Docker", "MCP"], "impact": "v2.0.9 released. Production-grade, fully tested, self-hostable in one command.", "link": "https://github.com/achref-soua/helio", "linkLabel": "View on GitHub" }, { "title": "Galley", "featured": true, "category": "Open Source", "range": "2025 – Present", "role": "Author", "summary": "Local-first LaTeX studio built with Tauri 2 and Svelte 5. Sub-150ms compile feedback, sandboxed TeX execution, and zero cloud dependency.", "highlights": [ "Sandboxed shell: shell-escape guard, archive hardening, IPC-only renderer access.", "14,000+ Rust regions at 100% test coverage; 1,391 TypeScript tests.", "Real-time PDF preview, project-aware snippets, and CodeMirror 6 editor." ], "stack": ["Rust", "Tauri 2", "Svelte 5", "TypeScript", "CodeMirror 6", "LaTeX"], "impact": "v0.7.0 released. 100% Rust test coverage across 14,000+ regions.", "link": "https://github.com/achref-soua/galley", "linkLabel": "View on GitHub" }, { "title": "Pulse", "featured": true, "category": "LLM Systems", "range": "2025", "role": "Author", "summary": "Clinical decision support for endovascular surgery. Full-stack app (Next.js 15 + FastAPI) with hybrid RAG over Quiver and a streaming multi-agent copilot via Groq.", "highlights": [ "Hybrid RAG (dense + sparse + reranking) over Quiver for surgical knowledge retrieval.", "LangGraph multi-agent copilot streamed via Groq SSE for real-time clinical Q&A.", "6 validated clinical scoring models for risk stratification and stent sizing.", "IFU compatibility engine for stent selection + auto-generated PDF patient reports." ], "stack": ["Next.js 15", "FastAPI", "Python", "LangGraph", "Groq", "Quiver", "RAG", "ONNX"], "impact": "End-to-end pipeline: patient data → clinical scores → stent recommendation → PDF report.", "link": "https://github.com/achref-soua/pulse", "linkLabel": "View on GitHub" }, { "title": "AI Resume Tailor", "category": "LLM Systems", "range": "2024", "role": "Lead Developer", "summary": "Adapts a resume to a target job description using LLMs, then exports an ATS-friendly PDF. Cuts tailoring time from hours to minutes.", "highlights": [ "OpenAI and Groq API integration for context-aware resume rewriting.", "ReportLab PDF generation with clean, ATS-compatible formatting.", "Iterative editing loop for fast, targeted application material." ], "stack": ["Streamlit", "Groq API", "OpenAI", "ReportLab", "Prompt Engineering"], "impact": "Resume tailoring time: hours → minutes.", "link": "https://github.com/achref-soua/Streamlit-resume-tool-app-using-Groq-LLMs", "linkLabel": "View on GitHub" }, { "title": "FinOps Multi-Agent", "category": "LLM Systems", "range": "2025", "role": "LLM Engineer", "summary": "Modular LangGraph assistant for cloud cost analysis, forecasting, and optimization. Deployed serverless on AWS.", "highlights": [ "Separate agents for cost analysis, forecasting, and recommendation tasks.", "+23% dialogue coherence via custom long-term memory for multi-turn conversations.", "Serverless deployment on AWS Lambda, S3, and Bedrock with CI/CD." ], "stack": ["LangGraph", "LLaMA 3", "AWS Lambda", "AWS Bedrock", "FastAPI"], "impact": "Cloud cost analysis time: hours → minutes.", "link": "https://github.com/achref-soua/AWS-FinOps-assistant", "linkLabel": "View on GitHub" }, { "title": "Cardiac MRI Diagnosis", "category": "Medical AI", "range": "2023", "role": "Team Lead", "summary": "Competition-winning cardiac MRI pipeline: segmentation, radiomics extraction, and stacked ensemble classification.", "highlights": [ "U-Net, SegNet, and SAM for cardiac structure segmentation.", "100+ radiomics features extracted with PyRadiomics.", "LightGBM + XGBoost + Autoencoder-SVM stacked ensemble." ], "stack": ["U-Net", "SAM", "PyRadiomics", "LightGBM", "XGBoost", "Medical Imaging"], "impact": "1st place among 100 participants. 95% train accuracy, 81% validation.", "link": "https://github.com/achref-soua/AI-assisted-cardiac-MRI-diagnosis", "linkLabel": "View on GitHub" }, { "title": "Satellite Algal Bloom Detection", "category": "Research", "range": "2022 – 2023", "role": "Solo Competitor", "summary": "Satellite imagery pipeline for harmful algal bloom detection using multi-spectral Sentinel-2 and Landsat data with optimized gradient boosting.", "highlights": [ "Multi-spectral and temporal feature engineering across diverse geographic regions.", "LightGBM hyperparameter search with Optuna." ], "stack": ["Remote Sensing", "LightGBM", "Optuna", "Sentinel-2", "Python"], "impact": "Top 4% — 50th out of 1,377 participants. AggRMSE 1.0536.", "link": "https://www.drivendata.org/competitions/143/tick-tick-bloom/", "linkLabel": "View Competition" }, { "title": "Edge AI Gait Analyzer", "category": "Research", "range": "2023 – 2024", "role": "Research Engineer", "summary": "Knee angle prediction from IMU sensors on edge hardware, targeting rehabilitation and bionic leg control. Resulted in a published paper.", "highlights": [ "Arduino Nano 33 IoT + MPU-9250 for motion capture.", "Denoising autoencoders improved signal quality by ~20%.", "Benchmarked GRU, LSTM, and Transformer architectures for time series prediction." ], "stack": ["Arduino", "IMU", "LSTM", "GRU", "Transformers", "Autoencoders", "Python"], "impact": "Published at GEEE 2024. Edge-ready AI for bionic rehabilitation.", "link": "https://cnpsi.com/PBS/Vol12/GEEE-12.pdf", "linkLabel": "View Publication" }, { "title": "Adaptive Nutrition Planner", "category": "Research", "range": "2024", "role": "Lead Engineer", "summary": "Adaptive meal recommendation combining Q-learning and genetic algorithms for constraint-aware meal selection.", "highlights": [ "Q-learning + genetic algorithms for meal selection under allergy and macro constraints.", "Feedback loops for dietary preferences and macronutrient targets.", "Nutritional data structured for 200+ meals." ], "stack": ["Python", "Q-Learning", "Genetic Algorithms", "Reinforcement Learning"], "impact": "Constraint-aware meal recommendations that adapt to user feedback.", "link": "https://github.com/achref-soua/Meal-recommendation-system-using-Q-learning", "linkLabel": "View on GitHub" } ]