About Me
I'magradstudentatArizonaStateUniversitypursuingmyMSinComputerSciencewithaperfect4.0GPA.IdidmyundergradinAI&DataSciencefromtheUniversityofMumbai,andsincethenI'vebeenhookedonbuildingthingsthatsitattheintersectionofsoftwareengineeringandmachinelearning.
Currently,IworkasaResearchAideatASU,whereIbuildproductionFastAPImicroserviceswithRAGpipelinesusingLangChainandFAISS.Outsideofwork,Iloveshippingsideprojects—fromareferralplatformwith1,800+userstoanML-poweredwaterfootprinttrackerdownloadedin50+countries.Igetakickoutoftakinganideafromzerotoproductionandseeingrealpeopleuseit.
MS in Computer Science
Arizona State University
GPA: 4.0/4.0
Based in Tempe, AZ
Originally from Mumbai, India
Research Aide @ ASU
Building RAG pipelines & microservices
Full-Stack + AI/ML
From Flutter apps to ML models
GitHub Contributions
Technical Skills
Experience
- Built production FastAPI microservices with RAG pipeline using LangChain and FAISS vector database, deployed on AWS Lambda with CI/CD via GitHub Actions to handle 500+ concurrent queries daily
- Engineered chat-history retrieval system with PostgreSQL backend implementing connection pooling and Redis caching layer, reducing average query latency from 800ms to 320ms (60% improvement)
- Created cross-platform mobile application using Flutter and Supabase, implementing OAuth 2.0 authentication with Google Sign-In and real-time data synchronization serving 300+ users across rural India
- Designed accessible UI with multi-language support (English, Hindi, Marathi) and voice navigation using Flutter TTS/STT, reducing onboarding completion time from 8 minutes to 5 minutes (40% improvement)
- Developed RESTful APIs for ERP system using Django REST Framework with Redis caching and Celery for async task processing, handling 50K+ daily requests with average response time of 60ms
- Optimized PostgreSQL queries by adding composite indices on frequently-joined columns and refactoring Django ORM queries to use select_related/prefetch_related, reducing report generation time from 45 seconds to 6 seconds
Featured Projects
ADAPT-SQL
State-of-the-art Text-to-SQL system achieving 93.7% execution accuracy on the Spider benchmark using fully local LLM processing, outperforming GPT-4 based approaches like DAIL-SQL and DIN-SQL.
Technical Deep-Dive
11-step pipeline spanning schema linking, complexity classification, similarity search, adaptive SQL generation, and validation-feedback retry. Achieves 93.7% execution accuracy on Spider 1.0 using fully local Qwen3-Coder via Ollama — no API costs, outperforming GPT-4 based DAIL-SQL (86.6%) by 7.1 points.
Referrlyy
Referral networking platform connecting job seekers with employees at top companies. Built cross-platform mobile app and web portal for seamless referral management.
Technical Deep-Dive
Built end-to-end referral workflow with PostgreSQL for data persistence, Node.js/Express for API layer with request validation middleware, and Flutter using BLoC pattern for state management. Deployed on Render with automated CI/CD pipelines.
BoltPrep
AI-powered mock interview app with real-time speech-to-text, intelligent answer evaluation, and personalized feedback to help users ace their interviews.
Technical Deep-Dive
Engineered low-latency audio pipeline combining Deepgram streaming STT with Gemini API for contextual answer evaluation. Used Riverpod for state management with caching strategies to minimize API calls while maintaining real-time responsiveness.
CaseTrack
WiCS × OpHack 2026AI-powered case management platform for nonprofits — voice-to-notes, photo-to-intake, semantic search, and automated funder report generation across 99 languages.
Technical Deep-Dive
Integrated Google Gemini 2.5 Flash for document processing and report generation, ElevenLabs Scribe for voice transcription, and pgvector for semantic case search. Row-level security across 13 tables ensures complete multi-tenant data isolation.
Aqua Trace
ML-powered water footprint calculator that uses image recognition to identify food items and track daily water consumption with environmental impact insights.
Technical Deep-Dive
Applied transfer learning on MobileNet V2 with custom classification head, using data augmentation techniques to achieve 92% accuracy. Converted model to TFLite with quantization for efficient mobile inference under 100ms.
Waste2Wealth
LA Hacks 2026Turn trash into cash — report litter, clean it up, get paid in Solana. A full 3-sided crypto economy with AI photo verification and World ID fraud prevention, built in 36 hours.
Technical Deep-Dive
Built a full blockchain payment loop on Solana devnet — reporter photographs garbage, AI verifies authenticity via Cloudinary, cleaner uploads after photo, community verifies, and all three parties receive automatic SOL payments. Entire cycle runs in under 2 minutes.
Mili
AI mental health companion built at SunHacks 2024 featuring voice-enabled conversations, mood tracking, and personalized wellness insights.
Technical Deep-Dive
Designed efficient context management by chunking conversations into summaries before LLM calls, reducing token consumption by 60% while preserving conversational continuity. Built mood tracking with PostgreSQL time-series queries for trend analysis.
Auto EDA
Automated exploratory data analysis tool that generates comprehensive visualizations, statistical summaries, and insights from any dataset with minimal configuration.
Technical Deep-Dive
Built modular analysis pipeline using Pandas for data profiling, automatically detecting numeric/categorical columns and applying appropriate statistical tests. Implemented Streamlit caching for efficient handling of large datasets.