Danish AhmedAI Engineer

Problem

Existing recruitment and knowledge-reasoning workflows relied on manual processing, simple rule-based logic, and slow AI pipelines, resulting in high latency, low transparency, and limited scalability across enterprise use cases.

Approach

Designed and optimized end-to-end AI pipelines for resume parsing, candidate ranking, and automation using hybrid LLM + OCR architectures. Built modular, agent-based systems for parsing, scoring, ranking, reporting, and outreach. Implemented custom ranking pipelines with transparent justification via FastAPI and Python ETL workflows. Developed advanced RAG and graph-based reasoning engines for large-scale text corpora, combining knowledge graphs, iterative query refinement, and prompt optimization. Developed LLM applications using LangChain, LangGraph, and Hugging Face, implementing RAG systems with ChromaDB for context-aware responses and multi-agent workflows. Built agentic AI systems with autonomous decision-making, enabling task decomposition, tool use, and self-correction through iterative reasoning with open-source models via Ollama. Automated orchestration and monitoring using n8n and LangSmith.

Impact

Reduced resume parsing latency from 15 seconds to 3 seconds through parallel processing and batching while maintaining high accuracy. Achieved candidate scoring at scale, reduced manual data wrangling by 80%, and delivered AI systems with over 90% retrieval and reasoning accuracy. Enabled enterprise clients to deploy transparent, scalable, and auditable AI-driven workflows across recruitment and knowledge reasoning domains.

Technologies

Problem

An internal application relied on hard-coded rules to analyze structured input data, making it difficult to adapt when data patterns changed and limiting personalization capabilities.

Approach

Collected and cleaned tabular datasets using Pandas, engineered input features, and trained classification models using Scikit-learn. Developed recommendation systems using collaborative filtering (SVD, ALS) and content-based filtering (TF-IDF, cosine similarity). Built classification pipelines with Random Forest and XGBoost on imbalanced datasets, applying SMOTE for data augmentation and hyperparameter tuning via GridSearchCV. Implemented a PyTorch prototype to compare results against classical models. Wrapped selected models inside FastAPI services with `/predict` endpoints and validated requests using structured payloads. Managed code versions and iterations using Git.

Impact

Replaced rule-based logic with deployable ML inference services, enabling model-driven decision-making and faster iteration cycles. Improved user engagement by 23% and click-through rate by 20% through recommendation systems. Achieved 92% precision on classification tasks involving imbalanced data, increasing reliability of predictions in production scenarios.

Technologies

Problem

Students struggled to translate theoretical concepts in Python and machine learning into working code for data analysis and modeling tasks.

Approach

Mentored 120 students in Python programming labs through hands-on exercises. Taught implementation of algorithms related to linear algebra, calculus, and statistics in practical coding contexts. Conducted guided lab sessions covering data preprocessing, data visualization, and implementation of basic machine learning models. Reviewed student code, debugged errors, and provided targeted feedback to improve correctness and clarity.

Impact

Enabled students to independently complete data analysis and machine learning assignments, improving code quality, conceptual understanding, and confidence in working with real datasets. Supported large-scale instruction that increased student engagement and hands-on proficiency in quantitative programming.

Technologies

Three-course specialization taught by Andrew Ng covering regression, classification, neural networks, decision trees, recommender systems, and best practices for ML development. Includes hands-on projects implementing algorithms from scratch.

Focused on building the mathematical foundations required for machine learning, including linear algebra, calculus, and probability & statistics. Included practical exercises applying math concepts to real ML problems.

Built a comprehensive time-series forecasting pipeline to predict hourly energy demand using statistical and deep learning models.

Problem

Accurate energy demand forecasting is crucial for grid stability and cost optimization.

Approach

Implemented ARIMA, Prophet, and LSTM on 35K hourly records (2015–2018). Created automated backtesting and a Streamlit dashboard for comparison.

Impact

Achieved 1% MAPE with LSTM, reducing forecasting error by 83%.

Technologies

Developed a secure digital voting system using a custom blockchain and cryptographic techniques.

Problem

Traditional voting systems face transparency and tampering concerns.

Approach

Implemented Python-based blockchain with homomorphic encryption for privacy-preserving vote validation.

Impact

Enabled verifiable and auditable elections while preserving voter anonymity.

Technologies

Performed survival analysis on censored transplant patient data to identify outcome predictors.

Problem

Highly censored medical datasets complicate outcome prediction.

Approach

Feature engineering and model comparison using XGBoost, LightGBM, and CatBoost.

Impact

Achieved 0.79 AUC in patient outcome prediction.

Technologies

Designed a navigation robot with computer vision and multi-sensor fusion.

Problem

Autonomous navigation requires reliable real-time obstacle detection.

Approach

Used YOLOv5 on Jetson Nano with ultrasonic and IR sensors integrated via Arduino and ESP32.

Impact

Enabled dynamic speed control and accurate point-to-point navigation.

Technologies

Built a predictive system to recommend top crops based on soil and climate data.

Problem

Farmers struggle to select crops that maximize yield.

Approach

Performed cleaning, merging, and regression modeling on 30 years of data.

Impact

Achieved 0.84 R² and delivered top-5 crop predictions.

Technologies

Built smart gloves capturing hand motion and transmitting data for ML classification.

Problem

Communication barriers for sign language users.

Approach

Used flex sensors and IMU sensors with ESP32 and cloud-based Random Forest.

Impact

Achieved 98% classification accuracy with real-time TTS output.

Technologies

Designed a low-cost smart home system with safety and automation features.

Problem

Smart home solutions are often expensive and complex.

Approach

Integrated ESP32 with gas, smoke, and thermal sensors plus irrigation automation.

Impact

Enabled no-code automation for end-users.

Technologies

Built CNC system to reproduce text and drawings with high precision.

Problem

Automated handwriting replication requires fine motor control.

Approach

Programmed G-code interpreter controlling stepper motors on a custom frame.

Impact

Achieved sub-millimeter drawing accuracy.

Technologies

Built a robot that adjusts speed based on real-time distance sensing.

Problem

Robots require adaptive navigation in cluttered spaces.

Approach

Used ultrasonic sensors and proportional control algorithms.

Impact

Enabled safe navigation in variable environments.

Technologies

Created a bot to retrieve and organize student results automatically.

Problem

Manual result collection is repetitive and slow.

Approach

Used Selenium WebDriver with structured storage logic.

Impact

Saved significant manual effort for bulk result retrieval.

Technologies

Analyzed Kaggle survey data for industry insights.

Problem

Understanding global ML trends requires deep data analysis.

Approach

Performed cleaning, augmentation, and visualization with Streamlit.

Impact

Identified salary and technology adoption patterns.

Technologies

Built a drone capable of autonomous person tracking.

Problem

Maintaining stable tracking requires robust vision algorithms.

Approach

Used Jetson Nano with Pixhawk and pixel-based tracking.

Impact

Achieved stable real-time following behavior.

Technologies