ANKITA NAIK
AI Engineer
Hyderabad,Telangana
Summary
AI Engineer specializing in real-time Computer Vision, LLM-powered RAG pipelines, Multi-Agent Systems, and Agentic AI workflows. Experienced in deploying production-grade deep learning models on NVIDIA edge hardware (TensorRT FP16) and building autonomous agent pipelines with LangGraph, tool-use, and self-healing orchestration. Proficient in RAG with FAISS/ChromaDB, LangChain ReAct agents, and multi-step reasoning. Published researcher in medical imaging.
Overview
3
3
years of professional experience
1
1
Certification
Work History
Junior AI Engineer
Retearn (Part of Recykal)
Hyderabad, Telangana
04.2025 - Current
- Anti-Pinch Vision System: Deployed real-time safety system with JSON-configurable polygon ROI segmentation; hybrid CV pipeline (contour-area deviation + pixel-wise background differencing) achieving 120 FPS on Jetson Nano.
- Edge Object Detection (RTMDet-S, TensorRT FP16): Production 5-class recycling classifier (mAP@0.5=0.984,
- Continuous RAG + Agentic AI Document Intelligence: Watchdog-driven delta-indexing into FAISS; LangChain ReAct agent with autonomous retrieval strategy, document version tracking, and grounded answer synthesis over live business documents.
- Deep Learning Image Retrieval: ViT-B/16 + FAISS IVF index for scalable brand similarity search; detection-based logo cropping pre-embedding; Flask REST API with incremental indexing.
- LSTM Predictive Maintenance & ETL Analytics: LSTM forecasting for machine anomaly detection; automated ETL log pipeline for large-scale device logs with Google Sheets API reporting.
Data Engineer Intern
Retearn (Part of Recykal)
Hyderabad, Telangana
09.2024 - 04.2025
- Object Detection & Data Pipelines: Designed and optimised production object detection pipeline on embedded hardware; built ETL pipelines and performed log analysis to drive performance improvements across the recycling classification system.
AI Research Intern
Centre of Excellence, SGGS&IT
Nanded, Maharashtra
07.2023 - 01.2024
- Retinal Disease Detection Research: Deep learning research for retinal disease AI; contributed to RFMiD 2.0 dataset publication in MDPI Data journal — now used globally as a benchmark.
Education
M.Tech - AI
SGGSIET
01.2023
B.Tech - E&TC
Dr. Babasaheb Ambedkar Technological University
01.2021
Skills
- Python
- SQL
- CUDA
- PyTorch
- TensorFlow
- Keras
- Scikit-learn
- OpenCV
- TensorRT
- ONNX
- LangChain
- LlamaIndex
- HuggingFace
- ChromaDB
- LangGraph
- RAG
- Agentic AI
- Multi-Agent
- ReAct Agents
- APScheduler
- FastAPI
- Flask
- Docker
- FAISS
- Prompt Engineering
- Vector DBs
- Object Detection
- Segmentation
- LSTM
- Transformer
- Jetson Nano
- Time-Series
- MMDetection
- DeepStream SDK
- GStreamer
- Git
- Linux
- Python programming
- Image processing
- Time management abilities
- Teamwork and collaboration
- Collaborative teamwork
- Multitasking Abilities
- Teamwork
Certification
Machine Learning with Python, Coursera, 2022-06-01
Publications
Retinal Fundus Multi-Disease Image Dataset (RFMiD) 2.0: A Dataset of Frequently and Rarely Identified Diseases., Data, 8, 2, 29, MDPI, https://www.mdpi.com/2098772
Projects
- WebTalker — AI Chrome Extension for Conversational Web Q&A
Built a privacy-first Chrome extension that lets users chat with any webpage instantly — zero cloud cost, zero data sharing.
How it works:
- Scrapes webpage content using BeautifulSoup4
- Converts content into embeddings using BAAI/bge-large (InstructorEmbeddings)
- Stores embeddings locally in ChromaDB vector database
- Answers questions using Groq LLaMA 3.3 70B at ~300 tokens/sec
Key highlights:
- MMR-based retrieval fetches diverse, relevant context — no repetitive answers
- ChromaDB disk persistence cuts repeat-visit load time from 30 seconds to under 2 seconds
- Flask REST backend with automatic re-indexing when page content changes
- Fully local pipeline — no third-party cloud embeddings needed
- Life Coach & Time Management AI Agent
A personal AI agent on Telegram that autonomously manages your daily life — finances, schedule, coaching, and reminders — all in one conversation.
Powered by a LangGraph ReAct agent with 4 specialised tool modules:
- memory_tool — remembers past conversations using ChromaDB vector memory
- finance_tool — tracks budget, income, and daily expenses
- schedule_tool — manages priorities, tasks, and time planning
- advice_tool — generates weekly reports and personalised coaching insights
Key highlights:
- Agent autonomously decides which tool to use based on your message
- Daily and scheduled reminders run in the background via APScheduler
- Groq LLaMA 3.3 70B for fast, zero-cost LLM inference
- Persistent memory across sessions — agent remembers you every time
- Fully self-hosted on local hardware — your data never leaves your machine
- Retinal Blood Vessel Segmentation & Multi-Disease Classification (M.Tech Project)
Deep learning research project to detect and classify retinal diseases from eye scan images — built as part of M.Tech thesis.
What it does:
- Segments blood vessels in retinal images using a CNN encoder-decoder architecture
- Classifies across 45 retinal disease categories
- Trained and evaluated on the RFMiD 2.0 benchmark dataset
Impact:
- Research formally published in MDPI Data journal — Data, 8(2), 29
- Dataset now used by researchers globally as a standard benchmark for retinal disease AI
- Contributed to advancing AI-driven diagnostic tools in medical imaging
Timeline
Junior AI Engineer
Retearn (Part of Recykal)
04.2025 - Current
Data Engineer Intern
Retearn (Part of Recykal)
09.2024 - 04.2025
AI Research Intern
Centre of Excellence, SGGS&IT
07.2023 - 01.2024
M.Tech - AI
SGGSIET
B.Tech - E&TC
Dr. Babasaheb Ambedkar Technological University