SS

Projects

Engineering case studies — each project is documented with the problem, architecture, challenges, benchmarks, and lessons learned.

Aster — Distributed AI Research Scientist

Built a hackathon-stage research orchestration MVP that turns an ML question into a literature brief, falsifiable hypothesis, bounded experiment plan, worker execution, analysis, and an evidence-labelled Markdown report.

Node.jsLLM APIsMulti-Agent SystemsPyTorch DDPgRPCDockerExperiment Orchestration
View Case Study
SYSTEM ARCHITECTUREDistributed AI research orchestrationResearch QuestionProblem intakeResearch PlannerQuestion-aware planLiterature RetrievalOpenAlex / Semantic ScholarResearch AgentsAnalyze evidenceExperiment PlannerMetrics + training planDistributed WorkersDMLF / PyTorch DDPResearch BriefEvidence + resultsOffline-demo mode uses deterministic agents and simulated workers where external services are unavailable.

Weather Prediction System

Built a city-specific weather prediction pipeline using XGBoost and historical weather observations, with preprocessing, lag-based feature engineering, and an application interface for accessing predictions.

PythonXGBoostPandasTime SeriesFastAPIDockerGradio
View Case Study
SYSTEM ARCHITECTURELarge-scale city-specific ML forecastingHistorical Weather90 GB datasetPreprocessingCleaning + feature engineeringCity-Specific ModelsThousands of locationsXGBoostPer-city forecastingFastAPIPrediction servingForecast OutputTemperature / weather predictionsCity-specific XGBoost models are exposed through an API-oriented serving layer.

Brain Tumor Segmentation

Investigated 3D deep-learning architectures for brain tumor segmentation using multimodal MRI data from the BraTS dataset.

PythonPyTorch3D U-NetAttention U-NetEfficientNetMONAI
View Case Study
SYSTEM ARCHITECTURE3D U-Net → EfficientNet → Attention pipelineBraTS 20203D MRI volumesPreprocessingNormalization + patchesBaseline 3D U-NetDice 0.8057EfficientNet U-NetDice 0.8157Attention + EfficientNetDice 0.8256ValidationDice + segmentation metricsBest validation Dice: 0.8256 using the Attention + EfficientNet architecture.

Speech Emotion Recognition

Developed a speech emotion recognition pipeline using audio preprocessing, MFCC feature extraction, and deep-learning models for emotion classification.

PythonTensorFlowPyTorchLibrosaMFCCCNN
View Case Study
SYSTEM ARCHITECTUREAudio preprocessing + neural emotion classificationAudio InputSpeech waveformPreprocessingCleaning + augmentationMFCC / FeaturesTime-frequency representationCNNSpatial feature extractionBiLSTMTemporal modellingEmotion PredictionEmotion classesPublished research focuses on data augmentation and lightweight CNN-based speech emotion recognition.

GridLock — Traffic Demand Prediction

Built a traffic-demand prediction solution using gradient-boosting models, with experiments focused on validation design, distribution shift, and road-type-aware prediction.

PythonXGBoostCatBoostLightGBMFeature Engineering
View Case Study
SYSTEM ARCHITECTUREFeature engineering + gradient-boosting model comparisonerror analysisTraffic DataTime + road attributesFeature EngineeringTemporal + road-type featuresBoosting ModelsXGBoost / CatBoost / LightGBMError Analysis + BlendingRoad-type-aware experimentsDemand PredictionTraffic demand estimatesValidationHoldout error + distribution shiftModel selection and blending are evaluated against held-out data; no leaderboard score is assumed by this diagram.

AgentForge — Multi-Agent Software Engineering Assistant

Built a locally runnable multi-agent assistant that coordinates specialized agents to analyze tasks, propose software architecture, critique solutions, plan implementation, and synthesize a final response.

PythonOllamaLLMsMulti-Agent SystemsCode Generation
View Case Study
SYSTEM ARCHITECTURELocal multi-agent software engineering assistantSoftware RequestUser requirementsAnalystRequirements analysisArchitectSystem designCriticReview + validationPlannerImplementation planCode GenerationOllama / local LLMValidated ProjectCode + tests + fixesSpecialized agents collaborate locally with validation, caching, and code-generation workflows.

Distributed ML Training Framework

Developed an MVP for coordinating machine-learning training across heterogeneous computers, combining PyTorch Distributed Data Parallel with node registration, telemetry, and job scheduling.

PythonPyTorch DDPgRPCSQLiteDistributed SystemsScheduling
View Case Study
SYSTEM ARCHITECTURECluster management + PyTorch DDP worker planejob submissionstategRPCgRPClaunchlaunchtelemetrytelemetryUser / CLISubmit training jobCluster ManagergRPC + schedulerSQLite RegistryNodes + job stateNode Agent ATelemetry + hardware profileNode Agent BTelemetry + hardware profileDDP Worker APyTorch / NCCLDDP Worker BPyTorch / GlooHeterogeneous consumer nodes coordinated through the cluster-management layer.