Jun 2026 - Present | 📍 New York, NY

Working in the Artificial Intelligence for Vision Science (AI4VS) Lab (Department of Ophthalmology) on human-vision-inspired, interpretable diagnostic AI, across two projects:

• Eye-tracking simulation experiment: running an eye-tracking experiment that records visual attention during simulated image-reading tasks, toward interpretable, gaze-informed diagnostic models.
• Retinal-imaging classification: using Vision Transformers and diffusion models on trimester-labeled retinal images (Optos ultra-widefield and OCT), relating imaging and maternal history to hypertensive disorders of pregnancy, framing prediction across the preeclampsia spectrum (preeclampsia, eclampsia, superimposed preeclampsia), with an eye to transfer for adjacent cardiovascular questions such as ANOCA.

Stack: Python, PyTorch, Vision Transformers (ViT), diffusion models, medical imaging (Optos, OCT), REDCap

Jan 2026 - Present | 📍 New York, NY

Automated phenotype extraction for a 118-patient cardiac sarcoidosis cohort under IRB AAAV0341, transforming 50,486 Epic cardiology progress notes + 153 rheumatology consults into a structured 56-field clinical dataset for phenotyping and outcome modeling.

Highlights

• Reconstructed fragmented Epic notes into longitudinal per-patient timelines with visit-level delimiters.
• Built a two-pass de-identification workflow using regex and LLM fallback for names, DOBs, and MRNs.
• Used GPT-4.1 under Columbia IT HIPAA-compliant protocols with JSON-constrained extraction.
• Designed a 56-field schema covering demographics, symptoms, device history, imaging, labs, histopathology, and therapies.
• Added anti-hallucination safeguards to return Unknown or null when evidence was absent.
• Produced a 3-sheet Excel output with structured extraction, de-identification audit, and contamination checks.
• Validated a random sample through independent clinician chart review.

Stack: Python, pandas, OpenAI API, tiktoken, tqdm, openpyxl

Jan 2026 - Present | 📍 New York, NY

Built an automated Human Rights Due Diligence scoring framework for 27 defense manufacturers, grounded in UNGP, UNICEF CRBP, ABA Defense Industry HRDD Guidance, UN Six Grave Violations, and Arms Trade Treaty Article 7.4.

Highlights

• Designed a 5-dimension HRDD rubric covering policy commitment, risk assessment, prevention, monitoring, and remediation.
• Added child-rights scoring dimensions based on CRBP and the Six Grave Violations framework.
• Used a two-stage LLM workflow: evidence retrieval first, rubric-based scoring second.
• Preserved source URLs and quoted evidence for full auditability.
• Benchmarked against 12 previously human-rated companies.
• Evaluated reliability using weighted Cohen’s kappa, Krippendorff’s alpha, Spearman correlation, MAE, and confusion matrices.
• Reduced manual review effort by approximately 80% while keeping outputs auditable.

Stack: Claude API, Python, pandas, Excel reporting, statistical reliability metrics

Sep 2025 - Present | 📍 New York, NY

teaching = {
    "course": "Artificial Intelligence for Public Policy",
    "responsibilities": [
        "Grade assignments and exams",
        "Hold office hours and mentor students",
        "Support curriculum on AI ethics, governance, and policy applications"
    ]
}

Sep 2025 - Present | 📍 New York, NY

council = {
    "role": "Communications & Professional Resources",
    "responsibilities": [
        "Curate and share career resources, internship opportunities, and industry events for the Columbia MSDS community",
        "Coordinate communications between students, faculty, and external partners",
        "Design technical tools that improve the student experience end-to-end"
    ],
    "flagship_build": {
        "project": "DSI Course Evaluation Website",
        "link": "https://github.com/rishika1099/DSI-Course-Evaluation-Website",
        "description": "Student dashboard for Columbia MSDS course reviews with live Google Sheets data, personalized rankings, AI-summarized review deep dives, and course recommendations",
        "stack": ["Python", "Google Sheets API", "LLMs"]
    }
}

Jun 2026 - Aug 2026 | 📍 New York, NY

achievements = {
    "Child_Welfare_Risk_Modeling": {
        "description": "Developing predictive risk models on child welfare administrative data with explainable ML, fairness auditing, and causal adjustment",
        "tools": ["Python", "scikit-learn", "SQL", "NCANDS data"],
        "result": "Transparent decision-support modeling in a high-stakes public-sector setting 📈"
    }
}

Aug 2023 - Jul 2025 | 📍 Bengaluru, India

achievements = {
    "Financial_Forecasting": {
        "description": "Built gradient-boosted regression models in Databricks for financial forecasting across 12 business units",
        "tools": ["Databricks", "PySpark", "Power BI", "Power Apps"],
        "result": "Reduced forecast error by 23%; supported $100K+ annual cost optimization 💰"
    },
    "RPA_Automation": {
        "description": "Built production ETL pipelines and Blue Prism RPA bots with logging, retry logic, and exception handling",
        "tools": ["Blue Prism", "Python", "Selenium"],
        "result": "Reduced manual reporting effort by 85% and improved SLA compliance from 92% to 99% ⏱️"
    }
}

Jan 2023 - Jul 2023 | 📍 Hyderabad, India

achievements = {
    "Net_Zero_Emissions": {
        "description": "Built predictive analytics workflows on environmental data for emissions reduction",
        "tools": ["Python", "scikit-learn", "Alteryx", "Qlik Sense"],
        "result": "Supported emissions-reduction insights through analytics dashboards 🌱"
    },
    "NLP_Clinical_Trials": {
        "description": "Applied TF-IDF, NER, and text classification over clinical trial documentation",
        "tools": ["Python", "NLP"],
        "result": "Reduced manual review time for document analysis 📊"
    }
}

Feb 2022 - Mar 2022

achievements = {
    "Campaign_Data_Analysis": {
        "description": "Built Tableau dashboards to analyze campaign performance metrics",
        "tools": ["Python", "Tableau"],
        "result": "Improved campaign analysis and resource allocation 📈"
    }
}

Domains

Technical Areas

Project	Description	Tech Stack	Tags
Folio-Clinical-Multimodal-RAG	Multimodal medical record companion with PDF, photo, voice, and text ingestion; five-stage extraction pipeline; consensus extraction using embedding-cluster voting; longitudinal context injection from prior reports; 85.1% extraction micro-F1, 100% RAG recall@1, sub-2s median latency	FastAPI, MongoDB, Redis, React, Vite, Claude API
Federal-Eagle-AI-Legal-Assistant	Multi-agent CrewAI system for U.S. federal legal analysis with semantic retrieval over all 54 U.S. Code titles, precedent search, elements analysis, and draft generation; 87% precision@5 on benchmark queries	CrewAI, LangChain, ChromaDB, Streamlit
Just-Ask-Coach-Query-SQL-Translation	Natural language to SQL to visualization pipeline for sports performance analytics with semantic KPI retrieval, SQL generation, AST safety validation, self-verification, charting, and follow-up suggestions	FastAPI, SQLite, ChromaDB, Claude, React, Vite
Prescribed-Motion-Exercise-Recommendation-LLM	AI coaching system mapping natural language fitness queries to personalized exercise recommendations using two-stage retrieval over 100k+ exercises and Claude tool-use re-ranking	FastAPI, PostgreSQL, Supabase, Claude, Fly.io, Netlify
Ruchi-Pantry-to-Plate-Intelligence-Platform	AI food platform with video-to-recipe extraction, pantry-to-plate matching, nutrition and allergen data, health coaching, smart swaps, and adaptive meal plans	React, Vite, Framer Motion, OpenAI, Serverless
Reel-Chef-Video-To-Recipe-Extractor	Vision-language pipeline that converts cooking videos into structured recipes with ingredient lists, step-by-step instructions, timestamps, and estimated cook times	Python, Computer Vision, LLMs
Hey-Swiftie-Cluster-Emotion-Verse	AI diary that converts journal entries into personalized verses and music recommendations using DistilRoBERTa emotion classification, K-Means clustering, lyric embeddings, and RAG	Python, DistilRoBERTa, K-Means, OpenAI, React, RAG
DSI-Course-Evaluation-Website	Student dashboard for Columbia MSDS course reviews with live Google Sheets data, personalized course rankings, AI-summarized review deep dives, and course recommendations	Python, Google Sheets API, LLMs
AI-Blog-Assistant	GPT-4 blog generator with image generation and SEO optimization	OpenAI, Streamlit
Analogy-Tutor	Explains technical concepts through personalized analogies	OpenAI, Streamlit
Fake-News-Detection	TF-IDF and Linear SVM pipeline for news credibility prediction	TF-IDF, SVM

Project	Description	Tech Stack	Tags
Colon-Cancer-Trial-Causal-Analysis	Causal-inference re-analysis of the Moertel 1990 adjuvant colon cancer trial with five nested estimands: ATE estimation, backdoor adjustment, bad-control demonstration, CATE, mediation analysis, and transportability	Python, lifelines, statsmodels, EconML
Colorectal-Cancer-Risk-Analysis	Visual and statistical analysis of diet and lifestyle factors associated with colorectal cancer risk	R, ggplot2

Project	Description	Tech Stack	Tags
Safe-Start-NCANDS-Child-Welfare-Prediction	ML and predictive analytics framework for identifying high-risk child welfare cases using NCANDS administrative data, explainable models, and fairness auditing	Python, scikit-learn, SHAP, Jupyter
Diabetes-Risk-Prediction	Gradient boosting pipeline with SHAP explainability for diabetes risk prediction	XGBoost, SHAP
Heart-Disease-Prediction	Machine learning pipeline with EDA, SMOTE, and hyperparameter tuning for heart disease prediction	scikit-learn, SMOTE
Car-Price-Prediction	Ensemble-based car price prediction with SHAP feature explanations	XGBoost, SHAP
House-Price-Prediction	XGBoost regression model on California housing data with app deployment	XGBoost, Streamlit
Loan-Status-Prediction	Gradient boosting classifier for loan status prediction with SHAP explanations	XGBoost, SHAP
Red-Wine-Quality-Prediction	Gradient boosting regression with feature importance analysis for wine quality prediction	XGBoost, SHAP
Customer-Churn-Prediction	Predictive model for repeat purchase and churn behavior in food delivery	scikit-learn
Rock-Mine-Prediction	Sonar signal classification comparing multiple ML models	scikit-learn

Project	Description	Tech Stack	Tags
Kidney-Disorder-Detection	Deep learning system classifying CT scans for kidney disorders with 99.2% accuracy	TensorFlow, VGG19, ResNet50
Medical-Image-Analysis-Assistant	AI-powered medical image analysis assistant using Gemini Vision	Gemini, Streamlit
Cataract-Detection	CNN and transfer learning system for automated cataract detection	TensorFlow, CNN
Keratoconus-Detection	SVM and deep neural network models for keratoconus detection	SVM, DNN
Traffic-Sign-Classifier	VGG16 transfer learning model achieving 98% accuracy on GTSRB traffic sign classification	TensorFlow, VGG16
Plant-Disease-Detection	ResNet50 transfer learning model for 38 plant disease classes	TensorFlow, ResNet50

Project	Description	Tech Stack	Tags
KV-Cache-Implementation	Controlled single-codebase benchmark of five KV-cache optimization techniques on Llama-2-7B: KIVI 2/4-bit quantization, TopK sparse selection, SnapKV eviction, TransMLA latent projection, and KIVI×TopK hybrid. KIVI 4-bit achieved 1.93× decode throughput at batch size 32 with lossless LongBench quality	PyTorch, Hugging Face Transformers, Triton, Modal, W&B

Project	Description	Tech Stack	Tags
Android-Malware-Analysis	Deep learning malware detection using static and dynamic Android application features with 98% accuracy	Python, Deep Learning
Blockchain-Secure-Data-Storage	Secure medical data storage prototype using ECDSA signatures and Proof of Work	Blockchain, ECDSA
Modular-Image-Encryption	Custom image encryption framework using image slicing and key-based transformations for secure image storage and transmission	Python, OpenCV, NumPy

Expected Dec 2026 | New York, US

📚 Coursework

• 🧑‍🏫 Teaching Assistant for Artificial Intelligence for Public Policy
• 🔬 Research Assistant at Columbia Irving Medical Center and Columbia GSAS
• 🏛️ Data Science Institute Student Council: Communications & Professional Resources

May 2023 | Vellore, IN | CGPA: 9.51/10

• 🏆 Rank: 7/~200 (Top 4%)
• 🎖️ Recipient of Merit Scholarship 2019-2023
• 👥 Program Representative 2019-2023