Kilani Sai Nikhil
AI-Augmented Systems & Deep Tech / Edge ML Engineer
Technical Skills & Stack
Python 3.11, C++17, SQL, TypeScript, Dart 3.5, Linux Syscalls, POSIX Terminal, Bash
PostgreSQL, SQLite (AES-256 Encryption), Redis, Vector MBTiles, 3NF Normalization
PyTorch 2.5, YOLOv11-Nano, SAHI Slicing, LiteRT (Gemma 4 E2B), INT8 TFLite, Google Coral TPU
Deep JSCC Neural Mesh, Sentinel-1 SAR Radar, ROS 2 (rclpy), Docker Stacks, Firecrawl, MCP
Education & Academic Base
NxtWave Institute
B.Tech in Computer Science & Engineering
Verified Credentials & Honours
Executive Summary
AI-Augmented Systems and Deep Tech Engineer combining rigorous native computer science fundamentals (C++, Python, SQL relational design, Linux, Docker, PyTorch) with 10x agentic scaffolding velocity (Model Context Protocol, LiteRT on-device inference, semantic code graphs). Author of saara-ai (38 releases, 2,600+ PyPI downloads), architect of Vitt (100% on-device Gemma 4 E2B via LiteRT, AES-256 SQLite), trainer of AerialEye (1,900+ Hugging Face downloads, 89.4% mAP@0.5 YOLOv11), and author of Sleep Health Biometrics & Analytics (Kaggle Data Science Capstone).
Engineering Systems & Projects
Vitt (Artha)
Full-Stack Mobile App- Engineered a 100% private, on-device AI financial tracking mobile application running Gemma 4 E2B via LiteRT (~14 tokens/sec).
- Eliminated invasive SMS and contact permissions by extracting transaction alerts locally via Android Notification Listener.
- Designed local encrypted SQLite vault (AES-256) ensuring zero telemetry and zero cloud API expenditure.
- Fully compliant with India DPDP Act 2023 and SEBI AI Advisory standards; selected for NxtWave x OpenAI Academy x IndiaAI showcase.
saara-ai
Open-Source CLI & Python/NPM SDK- Architected an open-source local-first dataset synthesis and distillation engine published globally across 38 package releases.
- Generated 2,600+ direct PyPI downloads with active distribution on NPM (`npx saara-ai wizard`) and PyPI (`pip install saara-ai`).
- Orchestrated bounded autonomous ResearchAgents combining google-adk and crawl4ai to convert documentation into clean training datasets.
- Integrated dynamic local LLM routing (Ollama, vLLM) and a custom TUI curation console, exporting to Parquet, JSONL, and HF formats.
AerialEye
Computer Vision Model & Dataset- Fine-tuned YOLOv11-Nano with SAHI dynamic tile slicing (640x640) for small-object disaster detection from high drone altitudes.
- Curated and validated a unified dataset of 6,327 aerial images across 6 classes (human, sos, vehicle, flood, road_damage, crack).
- Achieved 89.4% mAP@0.5 validation accuracy with INT8 TFLite quantization for real-time inference on Google Coral Edge TPUs.
- Published model cards and pipelines on Hugging Face Hub (kilanisainikhil/AerialEye) with 1,900+ total community downloads.
Sleep Health Biometrics & Analytics
Data Science & Statistical Modeling- Engineered an exploratory data analysis (EDA) and feature engineering pipeline on multidimensional biometric datasets.
- Decomposed compound blood pressure readings into discrete systolic/diastolic components with normalized distributions.
- Constructed correlation heatmaps and violin plots modeling links between occupational stress, sleep architecture, and heart rate.
- Published reproducible Jupyter/Pandas workflows and analytical reports on Kaggle Hub for public peer review.