Los Angeles, CA

Ruchir
Jadhav.

Building resilient cloud systems and distributed backends. Passionate about AI/ML, systems engineering, and end-to-end software.

2+
Years exp.
6
Projects
3
Companies
Scroll

I am a Computer Science graduate student at the University of Southern California with experience across cloud infrastructure, backend and mobile systems, data engineering, and machine learning.

I currently work as a Software Development Engineer at Easley Dunn Productions, building the Node.js/Express backend for SafetyNet, a cross-platform React Native app, covering authentication, real-time messaging, and geospatial features.

Previously, I was a Software Development Engineer Intern at Amazon Web Services, where I built resilient backend systems for the Amazon EKS Metrics Service, and a Data Engineer at BusinessNext, building ETL pipelines and query-driven dashboards for a banking CRM.

My projects span real-time multimodal fact-checking, multimodal emotion detection, cooperative autonomous-vehicle intersection management, and database internals — including adding materialized views to DuckDB.

I am looking for graduate software engineering roles where I can apply strong systems, algorithms, backend, and full-stack engineering skills.

Contact

ruchirjadhav29@gmail.com

213-234-8896

Los Angeles, CA



Looking for

Graduate SWE roles — backend, cloud, distributed systems, AI/ML.

Available 2026

Feb 2026 – Present

Software Development Engineer

Easley Dunn Productions

Feb 2026 – Present
  • Built the Node.js/Express backend for SafetyNet, a cross-platform React Native app, implementing JWT + Firebase OTP authentication, a CTE-based social-graph discovery algorithm with Haversine proximity ranking, and three-tier rate limiting over PostgreSQL.
  • Engineered real-time messaging via Socket.IO with JWT-authenticated WebSocket handshakes, AES-256-GCM encrypted storage, and atomic PostgreSQL transactions for message persistence and notification fanout.
  • Shipped 46+ merged pull requests across backend and mobile, including a three-tier-fallback university autocomplete proxy, a Leaflet map WebView with geospatial filtering, and 40+ Firebase Analytics events across major user flows.

May 2025 – Aug 2025

Software Development Engineer Intern

Amazon Web Services (AWS)

May 2025 – Aug 2025
  • Built a four-component backfill system in Go for the Amazon EKS Metrics Service, resolving CloudWatch API outages that had caused up to 6-hour metric data gaps, with failure detection, hierarchical data preservation, and chronological replay for zero data loss.
  • Designed a hot-reloadable failure simulation framework with per-log-group injection rates (0–100%) and S3-synced JSON config for safe outage testing in beta, validated end-to-end via before/after CloudWatch dashboard comparisons.
  • Implemented offset-timed reprocess (30s) and republish (45s) services with AWS-specific retry/backoff, CloudWatch Logs batching, and S3 lifecycle tiering (Standard→IA→Glacier), shortening outage recovery time and ensuring continuous data availability.

Sep 2023 – May 2024

Data Engineer

BusinessNext

Sep 2023 – May 2024
  • Built and refined SQL queries on Linux using Django ORM to power client-portfolio dashboards for Relationship Advisors in BusinessNext's banking CRM, enabling the frontend team to deliver responsive reporting interfaces.
  • Developed data logic to surface product recommendation opportunities for banking clients, translating Relationship Advisor requirements into structured queries that identified better-fit financial offers from existing account data.
01

FACTUAL — Real-Time Multimodal Fact-Checking

  • Built a real-time speech fact-checking pipeline: Whisper ASR transcription, LLM-based claim extraction, and retrieval-augmented verification against live web evidence.
  • Constructed a 400-claim benchmark split across pre- and post-model-cutoff claims and converted to speech-like audio to stress-test the pipeline under realistic conditions.
  • Added a self-critique loop where a judge LLM evaluates evidence sufficiency and iteratively refines search queries, up to 10 iterations, to reduce hallucination.
  • Best configuration (GPT-4o, agentic) reached 84% accuracy and 0.83 F1 on post-cutoff claims, versus 63% accuracy for zero-shot prompting alone.
  • Prototyped the retrieval layer as a standalone tool (SearchAPI) using the Gemini API and Google Custom Search for article retrieval, summarization, and batch claim processing with structured JSON output.
PythonOpenAI WhisperGPT-4oGeminiRAGGoogle Search/News API
02

Multimodal Emotion Detection

  • Balanced the MELD dataset (2,000 train / 400 dev samples per class) via audio pitch/speed/noise augmentation and OpenCV-based video filters to correct severe class imbalance.
  • Fused three modalities — BERT (text, with dialogue context), HuBERT (audio embeddings), and OpenFace facial Action Units (video) — using weighted late fusion (50% text / 30% audio / 20% video).
  • Fine-tuned a LoRA-augmented LLM on facial Action Unit features to generate natural-language emotion explanations grounded in AU activations.
  • Weighted late fusion reached 60% accuracy and 38% macro F1, roughly double the 23% accuracy of early fusion; shipped a Streamlit app for end-to-end video-to-emotion inference.
PythonBERTHuBERTOpenFaceLoRAPyTorchStreamlit
03

Materialized Views in DuckDB

  • Extended DuckDB's parser, logical planner, and execution engine to support CREATE/REFRESH MATERIALIZED VIEW, with dependency tracking to auto-invalidate views when base tables change.
  • Stored materialized views in DuckDB's native columnar format so they inherit its indexing and compression, and modified the query planner to prefer them automatically when applicable.
  • Benchmarked against standard views on TPC-H queries: retrieval time dropped from ~0.3s to ~0.004s (up to 100x faster) at the cost of ~25% additional storage and higher one-time creation overhead.
  • Validated scalability up to 2TB datasets and confirmed safe concurrent refresh/query access via locking and transaction control.
C++SQLDuckDBTPC-H
04

CARLA Cooperative V2X Intersection

  • Designed a V2X-inspired virtual Intersection Manager that replaces traffic lights with an FCFS, conflict-aware reservation protocol for autonomous vehicles in CARLA.
  • Built a lightweight mailbox communication layer for REQUEST/PERMIT/CROSSING/CLEARED message exchange between the manager and per-vehicle agents, with geometry-based conflict-box detection for entry/exit.
  • Benchmarked against fixed-cycle traffic signals: 51% lower average crossing time (2.56s to 1.25s/vehicle), 105% higher throughput (48 vs. 23.4 cars/min), and 51% faster total queue clearance.
  • Logged arrival, permission, entry, and exit timestamps to CSV for reproducible performance analysis across synchronous simulation runs.
PythonCARLA 0.9.16V2X
05

Travel Recommendation & Route Optimizer

  • Scraped TripAdvisor via Selenium for the top ~120 rated spots across 5 Indian cities, geocoding each with GeoPy into a ratings/category dataset.
  • Fetched the user's live location and used the Google Maps API to compute real driving distance and duration to every catalogued spot.
  • Applied KNN to shortlist the 7 closest matching spots, then benchmarked Tabu Search, Genetic Algorithm, and Simulated Annealing to sequence an efficient visiting route.
  • Simulated Annealing found the shortest routes in testing, but Tabu Search converged 5-8x faster, making it the practical pick under time constraints; visualized routes with Folium/OpenStreetMap.
PythonSeleniumDjangoGeoPyFoliumGoogle Maps API
06

Sign Language Detection

  • Built a real-time gesture recognition system for accessibility-oriented use cases.
  • Collected and processed a diverse hand-sign dataset using feature extraction, normalization, and augmentation.
  • Improved model quality through hyperparameter tuning, transfer learning, dropout, and regularization.
  • Achieved 93%+ accuracy while maintaining low-latency inference.
PythonTensorFlowOpenCVREST API

Languages

PythonGoJavaScriptJavaCC++

Frameworks & Libraries

Node.js/ExpressReact NativeReactDjangoFlaskSocket.IOPyTorchTensorFlowScikit-LearnNumPyPandasOpenCVSeleniumBeautiful SoupHTMLCSSTailwind

Databases

PostgreSQLMySQLMongoDB

Tools, Cloud & DevOps

AWSCloudWatchS3DynamoDBFargateFirebaseGitLinux
Jun 2024 – Dec 2025

University of Southern California

Master of Science in Computer Science

Los Angeles, CA

GPA3.38

Coursework

Analysis of AlgorithmsMachine Learning for Data ScienceAdvanced Data StoresDeep LearningMultimodal Probabilistic LearningAutonomous Cyber-Physical Systems
Jul 2019 – Apr 2023

University of Mumbai

Bachelor of Technology in Computer Engineering

Mumbai, India

GPA9.03/ 10

Let's build
something
together.

I'm actively looking for graduate SWE roles in cloud, backend, distributed systems, and AI/ML. If you have an opportunity or just want to connect — reach out.

Ruchir Jadhav.

Built with Next.js · Tailwind · Framer Motion