Skip to content
View zzzzzzzzak's full-sized avatar

Block or report zzzzzzzzak

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
zzzzzzzzak/README.md

Siddiqui Zakir Ali

Senior Machine Learning Engineer • Multimodal Systems • Numerical Computing
📍 Global / Remote • zak.siddiqui16@gmail.com • GitHub: @zzzzzzzzak • Live Portfolio


Executive Profile

Senior Machine Learning Engineer specializing in model development, high-throughput inference systems, numerical computing, and multimodal architectures. Experienced in architecting end-to-end production systems spanning low-precision quantization kernels, frontier LLM agent evaluation harnesses, dual-stream computer vision forensics, and dense video retrieval engines.

Core Specializations:
  ├── Deep Learning Systems (Quantization INT4/INT8, PEFT LoRA/QLoRA/(IA)³, Flash Softmax, GPU/CUDA)
  ├── Multimodal & Computer Vision (Dual-Stream 2D FFT Forensics, Video RAG, Temporal Grounding)
  ├── Marketplace & Platform Economics (Two-Sided Structural Elasticities, Dynamic Surge Pricing, Spatial Clearing)
  ├── Operations Research & Supply Chain (Multi-Echelon MEIO, Risk Pooling, Bullwhip Forensics, (s, S) Policies)
  ├── Econometrics & Marketing Mix Modeling (Weibull Adstock, Hill Saturation, SLSQP Counterfactual Reallocation)
  ├── Experimentation & Decision Science (CUPED Variance Reduction, Sequential SPRT, SRM Diagnostics)
  ├── Agent & LLM Evaluation (Statistical Hypothesis Testing, McNemar/Permutation Tests, CI Gates)
  └── Quantitative Modeling & Data Pipelines (Walk-Forward Backtesting, High-Throughput SQLite/Parquet)

Featured Flagship Repositories

Repository Domain & Architecture Core Tech Stack CI Status
marketplace-liquidity-pricing-engine Two-sided platform economics: Structural rider & driver elasticities ($D(p), S(p)$), welfare-maximizing dynamic surge multiplier optimization, spatial zone liquidity matching, and +44.6% GMV counterfactual lift. Python, SciPy, Statsmodels, Streamlit CI
supply-chain-inventory-optimizer Multi-echelon inventory optimization (MEIO): Multi-tier risk pooling across DC, Hub, and Retail tiers, stochastic lead-time safety stock ($z_\alpha \sigma_{\text{LTD}}$), (s, S) continuous review policies, and Bullwhip forensics. Python, SciPy, Statsmodels, Streamlit CI
saas-clv-retention-dynamics SaaS customer economics engine: BG/NBD latent transaction rates, Gamma-Gamma spend shrinkage, first-order absorbing Markov multi-touch attribution (removal effects), and Kaplan-Meier cohort survival curves with Greenwood 95% CIs. Python, SciPy, Statsmodels, Streamlit CI
causal-marketing-mix-modeling Production Causal Marketing Mix Modeling: Weibull lag adstock, Hill diminishing returns, Bayesian non-negative bounded ridge regression ($\beta_m \ge 0$), and SLSQP counterfactual budget reallocation with analytical gradients. Python, SciPy, Statsmodels, Streamlit CI
ab-testing-experimentation-platform Enterprise experimentation engine: CUPED variance reduction (41.5% sample savings), Wald's Sequential SPRT (6.3× peeking protection), Chi-square SRM diagnostic forensics, and Streamlit command center. Python, Statsmodels, SciPy, Streamlit CI
multimodal-peft-adapter-engine Modular Parameter-Efficient Fine-Tuning (PEFT): LoRA ($W_0 + \frac{\alpha}{r}BA$), QLoRA (NF4 4-bit NormalFloat + double quantization), (IA)³ activation scaling (<0.15% params), benchmarked VLM backbone. PyTorch, CUDA, PEFT, Streamlit CI
quant-ops-engine High-performance inference engine: INT4/INT8 bit-packing (7.47× compression), Flash Online Softmax, Chan-Golub-LeVeque LayerNorm, GradientHealthMonitor hooks. PyTorch 2.5, CUDA, NumPy CI
spatial-frequency-deepfake-detector Dual-stream deepfake forensics: SE-ResNet spatial boundaries + 2D FFT azimuthal power spectrum decay ($1/f^\alpha$), cross-attention gating, Grad-CAM attribution, video EMA. PyTorch, 2D FFT, OpenCV, Streamlit CI
multimodal-temporal-video-rag Dense video retrieval & temporal moment grounding: HSV/motion shot segmentation, contrastive dual-encoders, vector indexing, cross-modal attention re-ranking. PyTorch, Vector Search, OpenCV, Streamlit CI
llm-agent-eval-harness Production LLM & agent evaluation framework: 60-task golden benchmark (code, tool trajectories, QA anti-hallucination), paired permutation & McNemar hypothesis tests, CI regression gate CLI. Python, SciPy, Pytest, PyYAML CI
equity-forecasting-pipeline Quantitative equity pipeline: 120-ticker SQLite/Parquet warehouse, 10-window walk-forward backtest (RMSE + Sharpe ratio), automated 45-second morning briefing digest, Streamlit dashboard. Python, SQLite, Statsmodels, Parquet, Streamlit CI
customer-churn-retention-analysis Cost-sensitive customer churn modeling: 7,043 customer records, cost matrix optimization ($257k net retained value simulation), calibrated decision thresholding, Streamlit UI. Python, Scikit-learn, Pandas, Streamlit CI
enterprise-sales-bi-analytics Enterprise BI & Data Warehousing: Normalized Star Schema (40k fact records), SQL CTEs & window functions, Power BI DAX measure catalog, interactive Streamlit executive dashboard. Python, SQLite, SQL, Power BI DAX, Streamlit CI

Technical Competency Matrix

Machine Learning & Deep Learning:
  • Frameworks: PyTorch 2.x, TorchVision, Hugging Face Transformers, Scikit-learn
  • Parameter-Efficient Fine-Tuning (PEFT): LoRA (Low-Rank Decomposition), QLoRA (NF4 Double Quantization), (IA)³
  • Architecture Design: Vision Transformers, ResNet/CNN, Cross-Attention, Multi-modal Encoders, Dual Streams
  • Numerical Optimization: INT8/INT4 Bit-Packing, Shifted Two-Pass LayerNorm (Chan-Golub-LeVeque), Flash Softmax
  • Explainability: Grad-CAM, 2D FFT Azimuthal Power Spectrum, Cross-Modal Attention Heatmaps

Experimentation, Causal Inference & Decision Science:
  • Customer Lifetime Value: BG/NBD Latent Transaction & Dropout, Gamma-Gamma Spend Shrinkage, Discounted Cash Flows
  • Multi-Touch Attribution: First-Order Absorbing Markov Chains, Fundamental Matrix Removal Effects, Heuristic Audits
  • Survival & Retention: Kaplan-Meier Product-Limit Estimator, Greenwood Log-Log Confidence Bands, Log-Rank Tests
  • Econometric MMM: Weibull Lagged Adstock (PDF/CDF), Hill Diminishing Saturation Curves, Non-Negative Priors ($\beta_m \ge 0$)
  • Budget Optimization: SLSQP Non-Linear Constrained Reallocation, KKT Marginal ROAS Equalization (+1.09% Rev Lift)
  • Variance Reduction: CUPED (Univariate & Multivariate Pre-Experiment Covariates, OLS Residualization)
  • Continuous Monitoring: Wald's Sequential Probability Ratio Test (SPRT), Alpha-Spending Boundaries
  • Quality & Diagnostics: Sample Ratio Mismatch (SRM) Chi-Square Diagnostics, Multi-Dimensional Slicing
  • Hypothesis Testing: Welch's t-test, Mann-Whitney U, Bootstrap 95% CIs, FDR Benjamini-Hochberg

Retrieval, RAG & Vector Systems:
  • Video & Multimodal RAG: Temporal Shot Boundary Segmentation, Keyframe Attention Pooling, Temporal Grounding
  • Dense Retrieval: In-memory Dense Vector Indexing (Cosine / Inner Product), HNSW ANN principles
  • Evaluation: Recall@k, Temporal IoU (tIoU), mAP, Latency p50/p95/p99 Profiling

Evaluation & MLOps Infrastructure:
  • Frontier Agent Evals: Multi-turn Tool Trajectory F1, Sandboxed Code Execution, Anti-Hallucination Grounding
  • Statistical Rigor: Paired Permutation Tests, McNemar Significance Tests, Bootstrap 95% Confidence Intervals
  • CI/CD Automation: GitHub Actions workflows, automated regression gates, reproducible Pytest test suites

Marketplace Economics & Dynamic Pricing:
  • Structural Elasticities: Rider Price Decay Curves ($D_0 e^{-\epsilon_D (p-1)}$), Driver Recruitment ($S_0 (1 + \epsilon_S (p-1))$)
  • Dynamic Pricing: Welfare-Maximizing Non-Linear Surge Solvers, Regulatory Multiplier Caps, Price Gouging Protection
  • Spatial Liquidity: 24-Hour Diurnal Dispatch Simulation, Shock Event Forensics (Rainstorms, Morning/Evening Commutes)
  • Platform Economics: 75/25 Driver-Platform Revenue Splitting, Consumer Surplus Trade-Offs, Counterfactual GMV Lift (+44.6%)

Operations Research & Supply Chain Analytics:
  • Multi-Echelon Optimization: Multi-Tier Risk Pooling, Installation vs. Echelon Safety Stock, Capital Reduction
  • Policy Parameterization: Economic Order Quantity (EOQ), Continuous Review (s, S) and Periodic (R, s, S) Thresholds
  • Supply Chain Forensics: Multi-Tier Bullwhip Ratio Quantifications, Order Batching & Demand Variance Propagation
  • Catalog Analytics: ABC-XYZ Portfolio Segmentation, Service Level Frontier Trade-Off (85% to 99% CSL)

Data Engineering & Systems:
  • Languages: Python (Proficient), SQL (Advanced: CTEs, Windowing Functions, Index Optimization)
  • Data Warehousing: SQLite, Apache Parquet, Star Schema Dimensional Modeling
  • Production Dashboards: Streamlit Decision UIs, Power BI DAX Modeling

Contact & Collaboration

Popular repositories Loading

  1. matplotlib-tutorial matplotlib-tutorial Public

    Forked from rougier/matplotlib-tutorial

    Matplotlib tutorial for beginner

    Python

  2. ANN-Classification-Churn ANN-Classification-Churn Public

    Jupyter Notebook

  3. vinyl-records vinyl-records Public

    Sample Webpage for my first e-aommerce

    Python

  4. chess-application chess-application Public

    chess-openings

    TypeScript

  5. zzzzzzzzak.github.io zzzzzzzzak.github.io Public

    HTML

  6. customer-churn-retention-analysis customer-churn-retention-analysis Public

    End-to-end churn classification & retention ROI optimization across 7,043 subscriber records. Features automated validation, F1=0.84 ensemble, and Streamlit decision suite.

    Python