AI Engineer at ILMAscent (SocratifyMe)

Architected production multi-agent AI system using LangGraph for state-based orchestration of 6 specialized agents with conditional routing, cyclic workflows, and shared memory management to support adaptive educational feedback analysis and real-time pedagogical assessment generation. Built and deployed a retrieval-augmented generation (RAG) pipeline with 8 retrieval patterns (semantic search, multi-query expansion, hybrid dense-sparse retrieval, self-RAG with confidence thresholds, agentic retrieval with tools, corrective RAG with re-ranking) using OpenSearch Serverless, AWS Bedrock LLMs, and AWS Textract for multimodal document intelligence with OCR confidence scoring.

Applied AI Research Intern at Institute of Applied Artificial Intelligence and Robotics

Delivered color grading pipeline for emerald gemstones by curating and calibrating a 1,000+ image dataset with white-balance normalization, glare masking, and continuous LAB hue features, achieving 97% Random Forest CV accuracy and 95%+ SVM accuracy across bluish-green, pure green, and yellowish-green classes. Built an automated clarity assessment system with 37 initial images expanded to 1,000+ samples, applying CLAHE preprocessing, edge/texture features, and RF/SVM modeling.

Data Science Capstone Intern at Leadoff.ai

Integrated Hume.ai's multimodal sentiment analysis into a behavioral reasoning pipeline by engineering time-range aggregation functions that compute primary emotions and top-5 composite scores across four modalities (speech prosody, facial, vocal burst, language). Transformed text-based explanations into timestamped reasoning objects with start/end timestamps and speaker attribution, achieving 100% speaker classification accuracy (19/19) via fuzzy text matching and time-based alignment.

AI Software Engineer [Enterprise Solution] at Fractal Analytics

Prototyped a Generative AI-powered text-to-SQL solution using LangChain and LLMs; built a React-based chatbot to process natural-language queries and render dynamic tables, charts, and code; integrated Power BI dashboards for insights—boosting data engagement by 40% and presented to the CAIO and CTO. Designed GenAI prompts to auto-generate clear tables, KPIs, and definitions, cutting data interpretation time by 30% and increasing user satisfaction by 25%.

Data Engineer at Fractal Analytics

Built and optimized robust ETL/ELT pipelines using AWS (S3, Glue, Lambda, Airflow), SnapLogic, and Snowflake—boosting pipeline reliability and scaling client project revenue. Developed a reconciliation framework for BOTREE and SAP to automate discrepancy resolution and standardized across systems; improved Glue logging and SnapLogic API integration to enhance data sharing and reliability. Earned AWS Solutions Architect (SAA-C03) certification during this period.

Data Science Apprentice at Fractal Analytics

Engagement Prediction System: developed models including XGBoost, linear regression, and random forests; selected XGBoost for superior R² and RMSE scores, contributing to project delivery in mid-2022.

AI Engineer at ILMAscent (SocratifyMe)

Architected a production multi-agent AI system using LangGraph for state-based orchestration across six specialized agents, enabling adaptive educational feedback analysis and real-time pedagogical assessment generation. Built an 8-pattern RAG pipeline (semantic search, multi-query expansion, hybrid dense-sparse retrieval, self-RAG with confidence thresholds, agentic retrieval with tool use, and re-ranking) using OpenSearch Serverless, AWS Bedrock LLMs (Nova Pro, Claude Sonnet 4.5), and AWS Textract for multimodal document intelligence with OCR confidence scoring and math notation extraction. Developed MLOps infrastructure with automated evaluation (no human-in-the-loop), LangSmith observability, Lambda deployment via SAM/Docker, semantic caching (60% cost reduction), intelligent model routing, CI/CD via CodePipeline, distributed tracing (X-Ray/CloudWatch), prompt versioning with A/B testing, and security controls (IAM/KMS/Bedrock Guardrails) ensuring FERPA compliance and bias detectio

Applied AI Research Intern at Institute of Applied Artificial Intelligence and Robotics

Delivered a color grading pipeline for emerald gemstones by curating and calibrating a 1,000+ image dataset with white-balance normalization, glare masking, and continuous LAB hue features, achieving 97% Random Forest CV accuracy and 95%+ SVM accuracy across bluish-green, pure green, and yellowish-green classes. Built an automated clarity assessment system with 37→1,000+ samples, applying CLAHE preprocessing, edge/texture feature extraction, and ensemble modeling, reaching ~80% CV accuracy. Developed an end-to-end AI-powered gemstone grading pipeline combining classical CV (GrabCut, CLAHE) with deep learning (ResNet50 features, GAN augmentation) and Grad-CAM explainability.

Data Science Capstone Intern at Leadoff.ai

Integrated Hume.ai’s multimodal sentiment analysis into a behavioral reasoning pipeline by engineering time-range aggregation functions that compute primary emotions and top-5 composite scores across four modalities (speech prosody, facial, vocal burst, language). Transformed text-based explanations into timestamp-enriched reasoning objects with start/end timestamps and speaker attribution, achieving 100% speaker classification accuracy (19/19). Evaluated six multimodal sentiment platforms and designed a MongoDB schema for emotional intelligence storage (HumeProsody, HumeFace, HumeLanguage, HumeVocalBurst, HumeSpeakerComposite, HumeJobs).

AI Software Engineer (Enterprise Solution) at Fractal Analytics

Prototyped a Generative AI-powered text-to-SQL solution using LangChain and LLMs to enable NLP-based data queries and replace the homepage search. Built a React-based chatbot to process natural language queries and render dynamic tables, charts, and code; integrated Power BI dashboards for insights, boosting data engagement by 40%. Designed GenAI prompts to auto-generate clear tables, KPIs, and definitions, reducing data interpretation time by 30% and improving user satisfaction by 25%.

Data Engineer at Fractal Analytics

Built and optimized robust ETL/ELT pipelines using AWS (S3, Glue, Lambda, Airflow), SnapLogic, and Snowflake, boosting pipeline reliability and scaling client project revenue from $75K to nearly $500K through performance tuning, error handling, and real-time monitoring. Strengthened architecture by developing a reconciliation framework for BOTREE and SAP systems—automating discrepancy resolution and standardizing across systems; improved Glue logging and SnapLogic API integration. Earned AWS Solutions Architect certification (SAA-C03) and built pipelines using AWS services.