BlueCreast Engineering & Innovation Portfolios

Technical Showcases & Portfolios

Click any profile photo or card area below to view their dedicated engineering portfolio, case studies, and achievements.

Full Stack Engineering Showcase

Available for HireOpen

Prathmesh Lohar

Full Stack Web Developer & MERN Specialist

Engineering Student (SPPU, 7.40 CGPA). National Finalist in NEC 2025 at IIT Bombay, GeeksforGeeks Algorithm Awardee, and Ex-Junior Web Developer Intern at GWSWeb building responsive web apps and scalable platforms.

IIT Bombay NEC 2025 FinalistGFG DSA AwardEx-GWSWeb Intern

View Prathmesh's Portfolio

AI/ML & GenAI Systems Showcase

Available for AI RolesOpen

Yash Jondhale

AI/ML Engineer & GenAI Systems Architect

Computer Engineering Graduate (SPPU, 8.34 CGPA Distinction) specializing in LLM Orchestration, Retrieval-Augmented Generation (RAG), Computer Vision, and Scalable FastAPI Microservices.

Published in IJRTISPPU Distinction (8.34)Ex-Altair & Edunet

View Yash's Portfolio

Yash Jondhale

AI/ML Engineer • SPPU Distinction (8.34)

Published in IJRTI Journal

Ex-Intern @ Altair & Edunet

Founder & Lead AI EngineerSPPU Computer Engineering (8.34 CGPA Distinction)

Machine Learning Engineer, Published AI Researcher & GenAI Systems Architect

Computer Engineering Graduate (SPPU, 8.34 CGPA Distinction) specializing in LLM Orchestration, Retrieval-Augmented Generation (RAG), Computer Vision, and Scalable FastAPI Microservices.

Published in IJRTI (ISSN: 2456-3315)

Ex-Intern @ Altair Engineering & Edunet Foundation

8+ Production AI/ML & Computer Vision Systems

Founder & Lead Author @ BlueCreast

Download Resume (PDF)Contact Yash GitHub LinkedIn

AI & Machine Learning Engineering

Production AI Systems & Case Studies

RAG decision support, real-time computer vision telemetry, desktop automation, and predictive modeling.

Yash JondhaleGenAI & LLM SystemsFlagship

NyayaAI — Judiciary Legal Intelligence System

RAG-driven judicial decision-support, statutory fact extraction, and bail prediction with explainable AI (XAI)

Architected a production-ready RAG legal decision-support engine, automated statutory fact extraction, bail-prediction model, and an 'AI-as-Judge' module using Cosine Similarity, fine-tuned BERT embeddings, and Softmax/Sigmoid classification heads.

Prediction Accuracy

~85%

Security & Compliance

DPDP Act 2023 & SHA-256

Inference Latency

<850ms (Groq LPU)

Explainability

XAI Precedent Citations

PythonFastAPIRAGGemini APIGroq APIBERTPyTorchSQLOLAPREST APIs

Explore Case Study Source Code

ragEngine.pyFastAPI + BERT + Groq

# Hybrid Cosine + BM25 Legal Precedent Retrieval
@router.post("/api/v1/predict-bail")
async def evaluate_bail_risk(case_docket: CaseDocketSchema):
    doc_hash = hashlib.sha256(case_docket.text.encode()).hexdigest()
    query_vector = embed_model.encode(case_docket.facts)
    precedents = vector_store.similarity_search_by_vector(query_vector, k=5)
    bail_prob = classifier.predict_proba(query_vector)
    return {"doc_hash": doc_hash, "bail_prob": bail_prob, "precedents": precedents}

Engineering Highlights

•Cryptographic Chain-of-Custody: Document integrity preserved using SHA-256 fingerprinting, ensuring zero evidence tampering in alignment with India's DPDP Act 2023.
•Sub-Second LPU Execution: Integrated Groq API for inference acceleration, driving latency down from 4.2 seconds to 820ms on longform legal summaries.
•Transparent XAI Attribution: Replaced black-box LLM generation with strict citation anchoring, returning exact statutory sections and Supreme Court case numbers.

Lead: Yash JondhaleFull Case Study

Yash JondhaleGenAI & LLM SystemsFlagship

InterviewXpert — AI-Powered Mock Interview Platform

Dynamic resume-to-question LLM orchestration paired with real-time OpenCV computer vision for candidate behavioral telemetry

Built an LLM orchestration platform that parses candidate resumes to generate tailored technical interview questions and evaluates real-time video responses via OpenCV computer vision pipelines.

Video Analysis

Real-Time OpenCV Facial Telemetry

Question Generation

Dynamic Resume-to-Question Engine

Latency

<450ms WebSocket Audio/Video Pipeline

Evaluation

Multi-Dimension Candidate Scorecard

PythonReact.jsFastAPIGemini APIOpenCVLLM OrchestrationWebSocketsTailwindCSS

Explore Case Study Source Code

telemetryStream.pyFastAPI + OpenCV

# Real-Time Facial Landmark & Eye-Gaze Stream
async def process_video_frame(frame_bytes: bytes):
    np_img = np.frombuffer(frame_bytes, np.uint8)
    frame = cv2.imdecode(np_img, cv2.IMREAD_COLOR)
    landmarks = face_mesh.process(cv2.cvtColor(frame, cv2.COLOR_BGR2RGB))
    gaze_ratio = compute_gaze_drift(landmarks.multi_face_landmarks[0])
    return {"gaze_stability": gaze_ratio, "confidence": calculate_composure(frame)}

Engineering Highlights

•Dynamic Contextual Question Branching: If a candidate struggles with distributed systems, the system recursively prompts with diagnostic hints before pivoting to related architectural trade-offs.
•Edge-Optimized Gaze Tracking: Built low-compute OpenCV landmark algorithms that track pupil position without requiring heavy GPU acceleration on client hardware.
•Holistic Multi-Modal Scorecard: Synthesizes technical code accuracy, communication clarity (WPM and filler words), and emotional confidence into an actionable post-interview breakdown.

Lead: Yash JondhaleFull Case Study

Yash JondhaleGenAI & LLM SystemsFlagship

OMEGA — End-to-End AI Voice Assistant

Event-driven desktop voice assistant automating operating system commands, search pipelines, and productivity workflows

Engineered during the AI/ML Internship at Edunet Foundation (Bengaluru) to automate system commands, web searches, reminders, and weather pipelines via an event-driven PyQt6 interface.

Productivity Boost

40% Reduction in Manual Tasks

Deployment

Full Agile SDLC Deployment

Response Latency

<300ms Intent Classification

Operation

Hybrid Offline & Online Engine

PythonPyQt6SpeechRecognitionNLP Pipelinespyttsx3REST APIsAsyncIO

Explore Case Study Source Code

voiceDispatcher.pyPython + PyQt6 + AsyncIO

# Asynchronous Voice Command Intent Routing
class VoiceWorker(QThread):
    command_detected = pyqtSignal(dict)
    def run(self):
        with sr.Microphone() as source:
            audio = recognizer.listen(source, phrase_time_limit=5)
            transcript = recognizer.recognize_google(audio)
            intent = parse_system_intent(transcript)
            self.command_detected.emit(intent)

Engineering Highlights

•Zero-Cloud Offline Fallback: Designed to execute core operating system commands even in complete air-gapped environments without network connectivity.
•Non-Blocking Event Loops: Decoupled speech listening and audio synthesis from PyQt6 GUI rendering using QThread and AsyncIO workers, eliminating UI freezes.
•Extensible Plugin Architecture: Created a modular command registry pattern where new voice skills can be registered with a single Python decorator.

Lead: Yash JondhaleFull Case Study

Yash JondhalePredictive Data ScienceFlagship

Solar Power Generation Forecasting Engine

Predictive regression modeling and exploratory data science pipeline analyzing multi-sensor weather & irradiation telemetry

Engineered during the Data Science Internship at Altair Engineering Pvt. Ltd. (Pune) using Linear Regression, EDA, and historical weather/irradiation datasets to predict photovoltaic output.

Accuracy Gain

+10% Forecast R² Over Baseline

Interface

Interactive Streamlit Simulation UI

Data Pipeline

Automated Outlier & Missing Value Handling

Domain Impact

Grid Frequency & Battery Optimization

PythonScikit-learnPandasNumPySeabornStreamlitMatplotlib

Explore Case Study Source Code

solarForecaster.pyScikit-Learn + Streamlit

# Covariance Irradiation Telemetry Regression
def predict_generation(ambient_temp, module_temp, irradiation):
    features = np.array([[ambient_temp, module_temp, irradiation, irradiation**2]])
    predicted_kw = trained_pipeline.predict(scaler.transform(features))
    return float(np.clip(predicted_kw[0], 0, None))

Engineering Highlights

•Covariance Feature Modeling: Discovered non-linear inverse efficiency thresholds where extreme module temperatures degrade solar output despite high solar irradiation.
•Interactive What-If Dashboard: Empowered plant operators to simulate expected megawatt generation by manipulating expected cloud-cover percentages.
•Robust K-Fold Validation: Applied time-aware train-test splits to avoid temporal data leakage across seasonal transitions.

Lead: Yash JondhaleFull Case Study

Yash JondhaleComputer Vision & Deep LearningFlagship

Real-Time Plant Disease Detection System

Computer vision image classification pipeline utilizing fine-tuned CNN architectures for automated agricultural pathology

Designed and trained a Convolutional Neural Network (CNN) image analysis pipeline for early detection of localized plant leaf diseases across 38 crop disease classes.

Coverage

38+ Crop Disease Classes

Architecture

Deep CNN with Batch Normalization

Preprocessing

Adaptive Thresholding & Data Augmentation

Deployment

Edge-Ready Lightweight Inference

TensorFlowKerasCNNOpenCVPandasNumPyMatplotlib

Explore Case Study Source Code

pathologyModel.pyTensorFlow / Keras CNN

# Deep 38-Class Leaf Disease Classification
def classify_crop_pathology(image_tensor):
    normalized = tf.cast(image_tensor, tf.float32) / 255.0
    predictions = cnn_model.predict(tf.expand_dims(normalized, axis=0))
    class_idx = np.argmax(predictions[0])
    return {"disease": CLASS_LABELS[class_idx], "confidence": float(predictions[0][class_idx])}

Engineering Highlights

•Overfitting Immunity: Employed aggressive Dropout (0.5) and Batch Normalization layers to prevent memorization of laboratory background artifacts.
•Color Space Preservation: Retained essential leaf pigmentation cues by feeding full 3-channel RGB imagery with normalized floating-point tensors.
•Actionable Treatment Prescriptions: Mapped each disease class to an automated agronomic treatment guide, empowering farmers with immediate remedies.

Lead: Yash JondhaleFull Case Study

Yash JondhaleComputer Vision & Deep Learning

Garbage Classification System (CNN)

Automated municipal waste sorting using deep convolutional neural networks

Trained a high-precision multi-class classification model using TensorFlow and Keras to automatically categorize waste into recyclable, biodegradable, and hazardous streams for smart city recycling plants.

Classes

6 Waste Categories

Accuracy

92.3% Test Score

TensorFlowKerasCNNOpenCVPython

Explore Case Study Source Code

inferenceService.pyPython Microservice

# Production Inference Pipeline
def execute_pipeline(input_payload):
    sanitized = preprocess_features(input_payload)
    output = model.predict(sanitized)
    return format_results(output)

Lead: Yash JondhaleFull Case Study

Yash JondhaleGenAI & LLM Systems

Spam Mail Classification (NLP)

Natural language text classification engine for predatory message filtering

Built a high-throughput email spam filtering microservice utilizing TF-IDF n-gram vectorization and Naive Bayes / SVM classifiers, achieving 98.5% precision against phishing datasets.

Precision

98.5% Spam Precision

Latency

<15ms Per Email

PythonScikit-learnNLTKTF-IDFFastAPI

Explore Case Study Source Code

inferenceService.pyPython Microservice

# Production Inference Pipeline
def execute_pipeline(input_payload):
    sanitized = preprocess_features(input_payload)
    output = model.predict(sanitized)
    return format_results(output)

Lead: Yash JondhaleFull Case Study

Yash JondhalePredictive Data Science

E-Commerce Product Recommendation System

Hybrid collaborative filtering and content-based recommendation engine

Engineered an item-to-item and user-to-user collaborative filtering engine using sparse matrix factorization and Cosine Similarity over user transaction histories to maximize checkout cart conversion.

Recommendation Diversity

+34% Catalog Reach

Algorithm

Matrix Factorization (SVD)

PythonScikit-learnPandasNumPySurprise Library

Explore Case Study Source Code

inferenceService.pyPython Microservice

# Production Inference Pipeline
def execute_pipeline(input_payload):
    sanitized = preprocess_features(input_payload)
    output = model.predict(sanitized)
    return format_results(output)

Lead: Yash JondhaleFull Case Study

Peer-Reviewed Academic PublicationApril 2026 • Vol. 11, Issue 4

AI-Powered Legal Ecosystem: Advancing Justice Through Intelligent Adjudication, Professional Assistance, and Citizen Empowerment

Published in International Journal of Recent Technology and Engineering (IJRTI) (ISSN: 2456-3315)

This paper articulates an architectural framework for modernizing judicial workflows through artificial intelligence. By synthesizing dense semantic vector retrieval with retrieval-augmented generation (RAG) and explainable AI (XAI) classification heads, the proposed system demonstrates ~85% accuracy in statutory legal fact extraction and bail decision-support while strictly complying with the cryptographic and privacy mandates of India's Digital Personal Data Protection (DPDP) Act 2023.

Peer-Reviewed

IJRTI Certified

Paper ID

IJRTI2604038

Impact Factor

Indexed Journal

ISSN

2456-3315

Artificial IntelligenceRetrieval-Augmented Generation (RAG)Judicial Decision SupportExplainable AI (XAI)DPDP Act 2023Natural Language Processing

Verify Paper on IJRTI

Technical Capabilities

AI/ML & Engineering Skills Matrix

Validated technical competency across Large Language Models, Computer Vision, FastAPI microservices, and Predictive Modeling.

AI, GenAI & LLM Orchestration

Retrieval-Augmented Generation (RAG)Expert

LangChain & LlamaIndexAdvanced

Gemini API & Groq LPUAdvanced

Prompt Engineering & Few-ShotExpert

BERT & HuggingFace TransformersAdvanced

ChromaDB & FAISS Vector StoresAdvanced

Computer Vision & Deep Learning

PyTorch & PyTorch LightningAdvanced

TensorFlow & Keras CNNsAdvanced

OpenCV (cv2) Video TelemetryExpert

Image Segmentation & AugmentationAdvanced

Facial Landmark EstimationAdvanced

Backend Microservices & APIs

FastAPI & AsyncIOExpert

Python 3.11+Expert

WebSockets & Real-Time MediaAdvanced

Pydantic v2 & REST StandardsExpert

Uvicorn / Production DeploymentsAdvanced

Data Science, Analytics & Stats

Pandas, NumPy & SciPyExpert

Scikit-Learn Regression & TuningAdvanced

Streamlit Dashboard ArchitectureAdvanced

Statistical Hypothesis Testing & EDAAdvanced

Seaborn & MatplotlibAdvanced

Track Record & Applied Research

Work Experience & AI Leadership

Enterprise data science at Altair Engineering, production AI engineering at Edunet, and developer community contributor at GDG.

AI / Machine Learning Engineering Intern2024

Edunet Foundation

Bengaluru, India (Remote)

Spearheaded the research and end-to-end engineering of OMEGA, an intelligent system automation desktop voice assistant built on Python and PyQt6.

•Implemented real-time speech-to-text audio streaming using SpeechRecognition and low-latency offline speech synthesis via pyttsx3.

•Achieved a 40% reduction in manual desktop task time for developers through automated shell invocation, agenda scheduling, and web scrapers.

•Managed the full software development lifecycle (SDLC) under Agile sprints, completing design, testing, and documentation.

PythonPyQt6SpeechRecognitionNLP PipelinesSystem AutomationAgile SDLC

Data Science & Predictive Modeling Intern2024

Altair Engineering Pvt. Ltd.

Pune, India

Conducted extensive statistical data analysis, exploratory data science, and predictive machine learning on renewable energy datasets.

•Engineered a solar power generation predictive model utilizing multi-variate linear and polynomial regression, achieving a +10% improvement in forecast accuracy.

•Designed and deployed an interactive Streamlit simulation dashboard for renewable plant operators to model generation curves against weather variables.

•Cleaned, sanitized, and normalized large sensor telemetry time-series datasets using Pandas, NumPy, and Seaborn.

Data SciencePythonScikit-learnStreamlitRegression ModelingPandasEDA

Community Member & AI Technical Contributor2023 - Present

Google Developer Group (GDG Nashik)

Nashik, India

Active participant and technical contributor in community developer conferences, hackathons, and Google Cloud / GenAI developer sprints.

•Collaborated with industry practitioners on modern Gemini API integrations, LangChain orchestration, and scalable microservice architectures.

•Mentored student developers on Python best practices, git collaboration, and machine learning fundamentals.

GenAIGemini APICommunity MentorshipGoogle CloudHackathons

AI Prompting Event Head2024

Tech Guru Mega Event

Nashik, India

Headed the flagship AI Prompting competition for the regional collegiate technical symposium, orchestrating problem statements for over 200 participants.

•Curated complex multi-turn prompt engineering evaluation criteria testing context optimization, zero-shot/few-shot prompting, and hallucination reduction.

•Judged and graded technical submissions using automated scoring rubrics measuring factual accuracy and prompt compactness.

Prompt EngineeringLLM EvaluationEvent LeadershipTechnical Judging

Direct AI/ML Engineering Engagement

Looking for a Production-Grade AI/ML Engineer or GenAI Consultant?

Available for full-time Machine Learning roles, contract GenAI system architecture, RAG optimization, and AI microservice deployments. Let's discuss your technical roadmap.

📍 Nashik / Pune, Maharashtra•✉️ yashjondhale1294@gmail.com•🎓 SPPU B.E. Distinction (8.34 CGPA)

Download Resume Contact Yash Directly Connect on LinkedIn

Machine Learning Engineer, Published AI Researcher & GenAI Systems Architect

Computer Engineering Graduate (SPPU, 8.34 CGPA Distinction) specializing in LLM Orchestration, Retrieval-Augmented Generation (RAG), Computer Vision, and Scalable FastAPI Microservices.

Published in IJRTI (ISSN: 2456-3315)

Ex-Intern @ Altair Engineering & Edunet Foundation

8+ Production AI/ML & Computer Vision Systems

Founder & Lead Author @ BlueCreast

# Hybrid Cosine + BM25 Legal Precedent Retrieval @router.post("/api/v1/predict-bail") async def evaluate_bail_risk(case_docket: CaseDocketSchema): doc_hash = hashlib.sha256(case_docket.text.encode()).hexdigest() query_vector = embed_model.encode(case_docket.facts) precedents = vector_store.similarity_search_by_vector(query_vector, k=5) bail_prob = classifier.predict_proba(query_vector) return {"doc_hash": doc_hash, "bail_prob": bail_prob, "precedents": precedents}

# Real-Time Facial Landmark & Eye-Gaze Stream async def process_video_frame(frame_bytes: bytes): np_img = np.frombuffer(frame_bytes, np.uint8) frame = cv2.imdecode(np_img, cv2.IMREAD_COLOR) landmarks = face_mesh.process(cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)) gaze_ratio = compute_gaze_drift(landmarks.multi_face_landmarks[0]) return {"gaze_stability": gaze_ratio, "confidence": calculate_composure(frame)}

# Asynchronous Voice Command Intent Routing class VoiceWorker(QThread): command_detected = pyqtSignal(dict) def run(self): with sr.Microphone() as source: audio = recognizer.listen(source, phrase_time_limit=5) transcript = recognizer.recognize_google(audio) intent = parse_system_intent(transcript) self.command_detected.emit(intent)

# Covariance Irradiation Telemetry Regression def predict_generation(ambient_temp, module_temp, irradiation): features = np.array([[ambient_temp, module_temp, irradiation, irradiation**2]]) predicted_kw = trained_pipeline.predict(scaler.transform(features)) return float(np.clip(predicted_kw[0], 0, None))

# Deep 38-Class Leaf Disease Classification def classify_crop_pathology(image_tensor): normalized = tf.cast(image_tensor, tf.float32) / 255.0 predictions = cnn_model.predict(tf.expand_dims(normalized, axis=0)) class_idx = np.argmax(predictions[0]) return {"disease": CLASS_LABELS[class_idx], "confidence": float(predictions[0][class_idx])}