Sanket Shigaonkar

Software Engineer

Open to SWE/MLE/Data roles.
Learning Multi-Agent Systems & GraphRAG Architectures.
Open to Relocate | F-1 STEM OPT

My Resume

About Me

I am a Software Engineer currently pursuing my MS in Computer Science at SUNY Buffalo, driven by the challenge of making complex systems feel simple. My work sits right at the edge of AI and Backend Engineering, where I focus on turning experimental models into reliable, high-speed software.

Right now, I am deep into RAG pipelines and real-time data infrastructure. I love working across the entire stack—whether that means optimizing a database query for speed or fine-tuning an LLM for better accuracy. My goal is always the same: build tools that work flawlessly in the real world.

Previously, I engineered scalable solutions at NetCore Solutions and built privacy-focused synthetic data pipelines at Dentite, tackling everything from automated risk assessments to HIPAA-compliant AI systems.

In my spare time I am usually shooting hoops, watching the Warriors game (Steph is the GOAT), or unwinding to The Weeknd.

Some of the technologies I have been working on recently:

  • Python
  • RAG
  • LangChain
  • FastAPI
  • AWS
  • Kafka
  • Redis

Experience

  1. Software Engineer Intern · Dentite

    Jan 2025 — Jun 2025

    Buffalo, NY

    • Engineered a synthetic data generation pipeline creating thousands of insurance card images via SVG/PNG templates.
    • Optimized extraction accuracy by benchmarking and prompt-engineering multiple LLMs across 6 major US insurance providers.
    • Reduced model latency by 30% through optimization techniques.
    • Architected an automated claim reclamation feature that utilized extracted data to identify missed revenue, contributing to a 15% projected uplift for dental practices.
    • Python
    • LLMs
    • REST APIs
    • HIPAA Compliance
    • Data Pipelines
  2. Software Engineer · NetCore Solutions Pvt Ltd

    Jan 2023 — Jul 2024

    Mumbai, India

    • Drove a 10% efficiency gain in collections by engineering a multilingual intent classification pipeline (5 Indic languages) using OpenAI Whisper and Llama, optimizing repayment forecasting.
    • Automated credit risk assessment by configuring Experian PowerCurve strategies, eliminating manual underwriting for standard applications and ensuring real-time compliance.
    • Streamlined customer onboarding by implementing Business Rules Engine (BRE) workflows, reducing processing overhead and accelerating loan disbursement cycles.
    • Architected an internal Loan Management System (LMS), validating credit-bureau APIs with Postman/SoapUI for 100% integrity.
    • Python
    • OpenAI Whisper
    • Llama
    • Experian PowerCurve
    • Postman

Projects

  • Accio AI - Conversational RAG Agent

    • Engineered a Generative AI agent by fine-tuning Qwen-4B on 25,200+ samples via QLoRA-SFT for domain-specific reasoning.
    • Architected a sub-100ms RAG pipeline with FAISS and Flash Attention 2.
    • Achieved 77% accuracy and 56% loss reduction via RLHF alignment.
    • Qwen-4B
    • Hugging Face
    • FAISS
    • LangChain
    • Streamlit
  • Tickr - Live Crypto Price Monitor

    • Architected a real-time streaming platform with FastAPI and Kafka, ingesting high-frequency market data via Docker.
    • Engineered a sub-millisecond alert engine using Redis for O(1) lookups.
    • Deployed an event-driven AWS Lambda backend with DynamoDB for auto-scaling reliability.
    • FastAPI
    • Apache Kafka
    • Redis
    • AWS Lambda
    • Docker
    • DynamoDB
  • DeepDish - Deep Learning Calorie Estimator

    • Engineered a multi-view food recognition pipeline using YOLOv8 on ECUST Dataset, achieving 0.975 mAP on 19 classes.
    • Architected a volumetric engine using GrabCut and coin calibration to solve 2D-to-3D depth ambiguity.
    • Deployed a real-time Streamlit calorie tracker with less than 10% error.
    • Python
    • PyTorch
    • YOLOv8
    • OpenCV
    • Streamlit
  • Smart YOLO Intersection Traffic SystemPatent Filed

    Co-invented a real-time congestion controller utilizing YOLOv8 for vehicle detection and Residual LSTMs for predictive signal timing.

    • YOLOv8
    • Residual LSTM
    • Computer Vision
    • Deep Learning