Open to opportunities

Sakshi
Gangwani

AI Engineer & Full-Stack Developer

AI Engineer Intern at USC Alfred E. Mann School of Pharmacy and CS Master's student at USC Viterbi School of Engineering, building intelligent systems end-to-end from ML models and retrieval pipelines to the full-stack products around them. And yes, one of those projects even got a published patent with my name on it.

View ProjectsGitHubLinkedInResume
Sakshi Gangwani
// about me

A bit about
my journey

I'm Sakshi, a Master's in Computer Science student at the Viterbi School of Engineering, University of Southern California and an AI Engineer at the USC Alfred E. Mann School of Pharmacy and Pharmaceutical Sciences, building intelligent systems at the intersection of machine learning and full-stack engineering.

Before USC, I completed my B.Tech in Computer Engineering with Honors in Intelligent Computing from Dwarkadas J. Sanghvi College of Engineering, Mumbai. Along the way I led a full-stack DevOps team at a hardware startup, built industrial IoT systems, and published a patent as an inventor for an AI-driven gas leak detection robot.

That journey, from embedded firmware to ML pipelines, is what brought me here and what keeps me building.

3.5/4
USC GPA
1
Patent as inventor
Honors
Intelligent Computing
USC Viterbi School of Engineering
M.S. in Computer Science
University of Southern California · Los Angeles, CA · 3.5/4 GPA
Dwarkadas J. Sanghvi College of Engineering
B.Tech in Computer Engineering - Honors in Intelligent Computing
Mumbai, India
// experience

Where I've been building

USC
March 2026 - PresentLos Angeles, USA

AI Engineer Intern

USC Alfred E. Mann School of Pharmacy and Pharmaceutical Sciences · Dogra Lab

Built LLM-powered systems that automate scientific literature mining and structured data extraction for pharmaceutical R&D, spanning antibody, siRNA, and lipid-nanoparticle research.

  • Built a biomedical literature-mining pipeline that retrieved and screened 4,500+ candidate papers across 23 therapeutic antibodies down to ~430 structured study records, replacing the lab's manual review workflow (9 external APIs incl. PubMed; async rate-limited search; 3-tier DOI/PMID/fuzzy-title dedup).
  • Engineered a hybrid extraction engine pairing keyword heuristics with GPT-4o text + Vision (reading concentration–time figures) to auto-extract 20+ structured clinical fields per paper at ~95% accuracy - cutting search-and-extraction from ~1 day to ~10–15 minutes (~40× faster).
  • Shipped full-stack, self-serve extraction tools (FastAPI/Flask + React/TypeScript) enabling researchers to curate siRNA & lipid-nanoparticle datasets on demand, eliminating hours of ad-hoc spreadsheet work per dataset.
  • Eliminated fabricated LLM outputs by routing each field to deterministic computation, PubChem lookups, or cited LLM extraction with reference-level provenance - driving unverifiable values to ~0 while keeping every value source-traceable.
  • Applied vision models to parse data directly from scientific figures - charts, plots, and tables - extracting axis units, cohorts, and data points alongside 100+ structured text fields.
  • Engineered high-throughput screening with concurrent querying, disk-based caching, run persistence, and field-by-field Server-Sent Events streaming for responsive UX, plus PDF-native processing with OCR and figure extraction.
PythonFastAPIReactTypeScriptLLMsVision ModelsRAG / RetrievalServer-Sent EventsOCR
ME
November 2023 - January 2025Mumbai, India

Full Stack & Embedded Software Developer Intern

Meshnet Electronics LLP

Built memory-constrained embedded firmware and full-stack tooling for ESP32-based industrial IoT devices deployed across 100+ field units.

  • Optimized memory-constrained embedded C/C++ firmware for ESP32-based IIoT devices deployed across 100+ field units, with real-time task scheduling to run reliably under tight resource limits.
  • Hardened device-to-device communication with custom protocol stacks (Modbus, MQTT, OPC UA) and robust error-handling, ensuring reliable data transmission across noisy industrial environments.
  • Designed persistent-storage and fault-recovery mechanisms that guaranteed zero data loss during system failures.
  • Streamlined device deployment by 60% via real-time configuration interfaces on a C++ backend with a responsive HTML/CSS/JavaScript frontend.
C / C++ESP32ModbusMQTTOPC UAHTML/CSSJavaScriptEmbedded Systems
EC
September 2023 - November 2023New Jersey, USA

Software Developer Intern

Empire Circuits LLC

Developed desktop applications bridging antenna hardware and operators with real-time monitoring and control.

  • Developed desktop UI applications in C++/Qt (Qt Creator) with XML-based layouts, exposing 10+ configurable antenna parameters to operators.
  • Integrated real-time hardware-to-software data flow with responsive UI updates, enabling live monitoring of antenna performance during operation.
C++QtQt CreatorXMLReal-Time Systems
// patent

A patent with my name on it

During my undergrad, a project I co-built went all the way to a filed patent - and earned a formal Letter of Appreciation from the college.

Published Patent

IoT-Enabled Robot for Detecting Gas Pipe Leakage

Indian Patent Office · Application Number: 202421078054Published 15 Nov 2024verify status →

A mobile robot that autonomously navigates along gas pipelines and uses a machine-learning model to detect leaks from live multi-sensor telemetry - LPG concentration, temperature, and humidity. The work was filed as a patent and recognized with a formal Letter of Appreciation from the college Principal.

  • Designed an IoT inspection robot (Raspberry Pi + gas/temperature/humidity sensors) that streams live telemetry while moving through pipe networks.
  • Trained a K-Means clustering model (k = 3, k-means++) on IoT telemetry, using the elbow method to fix the optimal cluster count, to separate normal readings from leak signatures.
  • Flagged leaks in real time by mapping new sensor readings to the learned clusters - e.g. elevated LPG with characteristic humidity/temperature ranges triggers a "Gas Leakage Detected" alert.
  • Combined IoT, machine learning, and a companion mobile app into one end-to-end system for early, autonomous pipeline-leak detection.
IoTRaspberry PiPythonscikit-learnK-MeansMobile AppSensors
ML model on GitHubLetter of Appreciation
// projects

Things I've shipped

A selection of what I've been building. Click any card for the full story - the problem, what I did, and what came out of it.

// writing

Things I've written

Occasional notes on what I'm learning - the messy middle, not just the highlight reel.

MediumJun 20265 min read

RAG Explained: How AI Stops Making Things Up

A plain-English look at Retrieval-Augmented Generation - why language models hallucinate, how giving them a search step grounds their answers in real sources, and what makes RAG the backbone of trustworthy AI applications.

RAGLLMsRetrievalAI
Read on Medium
MediumJun 20264 min read

NLP: How Machines Learned to Read Between the Lines

A friendly tour through Natural Language Processing - how it grew from rigid rule-based systems to machine learning and finally the Transformer architecture that taught machines to actually understand the language we speak.

NLPTransformersDeep LearningAI
Read on Medium
MediumFeb 26, 20264 min read

The Algorithm Arc Nobody Saw Coming

How I went from dreading Data Structures & Algorithms to actually enjoying the grind - swapping outcome-anxiety for process, leaning on visualization, and letting consistency do the heavy lifting on the road to FAANG interviews.

DSAInterview PrepMindset
Read on Medium