Hello, I'm

Nitin
Jayavarapu

AI / ML Engineer · Open to full-time roles

I build production-grade AI systems — from computer vision pipelines and LLM agents to real-time data infrastructure. Currently at LumisAI, shipping models that handle millions of data points daily.

View Resume See My Work
0+
Years of Experience
0+
Projects Shipped
0%
Model Accuracy (peak)
0×
Avg. Speed Improvement
Nitin Jayavarapu
Fremont, CA · Open to work

About

Building AI that ships,
not just demos.

I'm an AI Engineer based in Fremont, CA. I specialize in computer vision, LLM-powered agents, and end-to-end ML pipelines. My work spans research and production — I care about systems that are accurate, fast, and maintainable.

I hold a Master's in Data Science from University of West Florida with a focus on AI/ML. In my experience, curating the right "hard examples" moves the needle faster than another round of hyperparameter tuning. This mindset shapes every system I build — from biometric pipelines to real-time detection APIs.

Projects

Selected work.

01

Data Cleaning Agent

An agentic AI system that autonomously detects and resolves data quality issues across diverse datasets. Combines LLM-powered reasoning with rule-based validators to identify missing values, schema inconsistencies, duplicates, and outliers — then applies targeted fixes with full audit trails.

↓ 60% manual effort GPT-4 + LangChain Multi-format support
LangChain OpenAI GPT-4 Pandas Python Agentic AI
→
The Challenge

Data quality issues are highly contextual — a missing value in one column might be fine, but critical in another. The core challenge was building an agent that reasons about intent rather than just applying rigid rules, while keeping fixes auditable and reversible.

View on GitHub
Key Highlights
  • Multi-tool LangChain agent with planner + executor pattern
  • Supports CSV, JSON, Parquet, and SQL table inputs
  • Generates human-readable fix reports with before/after diffs
  • Configurable validation rules via YAML — no code changes needed
  • Reduced analyst data prep time by 60% in testing
02

Resume Filtering Agent

An intelligent resume screening system that uses multi-agent AI to evaluate candidates against job descriptions. Extracts skills, experience, and qualifications using NLP, then ranks and filters applications with explainable scoring — reducing recruiter review time significantly.

↓ 70% screening time 92% precision Multi-agent pipeline
LangChain OpenAI spaCy FastAPI NLP
→
The Challenge

Recruiting tools often filter by keywords alone, missing strong candidates with non-standard experience. The challenge was building a system that understands equivalence — knowing "built ML pipelines" means the same as "developed machine learning workflows" — while keeping scores explainable to recruiters.

View on GitHub
Key Highlights
  • spaCy NER extracts skills, titles, years of experience automatically
  • Three-agent pipeline: parser → evaluator → ranker
  • Each score includes a natural-language explanation for recruiters
  • FastAPI backend accepts bulk PDF/DOCX uploads
  • Achieved 92% precision against a manually labeled test set of 500 resumes
03

MediBridge

A healthcare platform bridging patients and providers using AI-powered symptom assessment, appointment scheduling, and medical document summarization. Integrates with EHR systems to surface relevant patient history and flag critical information for clinicians.

↑ 40% triage efficiency HIPAA-aware design
React FastAPI OpenAI PostgreSQL LangChain
→
The Challenge

Healthcare data is sensitive and unstructured — patient notes, lab results, and discharge summaries all live in different formats. The challenge was summarizing and surfacing only the clinically relevant context at the right moment, without exposing unnecessary PHI or hallucinating medical facts.

View on GitHub
Key Highlights
  • Symptom checker uses a structured prompt chain to triage severity
  • Medical document summarizer built with retrieval-augmented generation
  • Role-based access: patients see summaries, clinicians see full context
  • All PII handled server-side; zero patient data sent to third-party APIs
  • React dashboard with real-time appointment availability via WebSockets
04

Smart Place Recommender

A context-aware recommendation system that suggests places — restaurants, parks, events — based on user behavior, real-time location, and preferences. Uses collaborative filtering combined with LLM-generated reasoning to produce personalized, explainable recommendations.

↑ 35% engagement Real-time recommendations Explainable AI
Python PyTorch Collaborative Filtering FastAPI Redis
→
The Challenge

Pure collaborative filtering suffers from cold-start problems and can't explain why it's recommending something. The challenge was combining the accuracy of matrix factorization with LLM reasoning to produce recommendations that feel personal and make intuitive sense to users.

View on GitHub
Key Highlights
  • PyTorch matrix factorization model trained on implicit feedback signals
  • LLM layer generates one-sentence reasoning per recommendation
  • Redis caches top-k results per user for sub-20ms API response
  • Geolocation filtering to prioritize nearby, open venues in real time
  • A/B tested against baseline: 35% lift in click-through engagement

Experience

Where I've worked.

AI Engineer Current
Jan 2025 — Present
LumisAI · Remote
  • Architected and deployed real-time computer vision pipelines processing millions of data points per day with sub-100ms latency
  • Built LLM-powered automation workflows that reduced manual processing time by 3× across core business operations
  • Led model optimization initiatives improving inference throughput by 60% using quantization and TensorRT
  • Integrated multi-modal AI systems combining vision, language, and structured data for enterprise clients
AI Engineer
Jan 2022 — Nov 2023
Black Box· Banglore, India
  • Engineered a face recognition system achieving 90% accuracy under high-occlusion (masked faces) and low-light conditions, significantly outperforming legacy baselines in production
  • Reduced model inference latency by 40% on edge devices through architectural optimization, enabling real-time processing for live biometric authentication
  • Applied data-centric AI principles — curated targeted "hard example" datasets that eliminated critical failure modes and improved model reliability during retraining cycles
  • Deployed production-grade REST APIs using FastAPI and Docker, integrating biometric authentication capabilities into mobile and web ecosystems

Skills

What I work with.

Computer Vision
PyTorch OpenCV YOLO TensorFlow TensorRT ONNX Image Segmentation Object Detection
LLMs & Agents
LangChain OpenAI API RAG Prompt Engineering Multi-Agent Systems Hugging Face Transformers
MLOps & Infrastructure
MLflow Weights & Biases Docker AWS FastAPI Apache Kafka GitHub Actions
Data & Frontend
Python SQL Pandas Spark PostgreSQL React Node.js

Get in Touch

Open to full-time roles.
Let's build something.

Looking for ML, CV, or AI Engineering roles. I reply to every message.

Email Me LinkedIn GitHub