About01Work02Projects03Experience04Toolkit05Contact06Resume

AI / ML Engineer · IIT Hyderabad

Ajay Wanekar

Perception → Reasoning → Action

©2026/Building since 2023

01 / About

Hey!

I'm Ajay, an AI/ML engineer and M.Tech student at IIT Hyderabad.

Portrait of Ajay Wanekar

I build perception and multimodal AI: from sonar and near-infrared object detection for autonomous vessels, to vision-language models, to LLM systems that retrieve, reason and respond.

Over 2+ years across three IIT Hyderabad labs, I've built datasets from scratch, fine-tuned VLMs and published my research at IEEE.

Now
M.Tech, Energy Science & Technology, IIT Hyderabad (2027)
Before
B.E. Electrical Engineering, SPPU (2022)
Labs
LFOVIA · TIHAN · WINET
Based in
Hyderabad, India
View resume

I engineer computer vision, vision-language and agentic AI, taking sensor data through fine-tuned models to systems that perceive, reason and act, and I lead that work from plan to delivery.

02 / Focus

What I work on

Open an area to see the projects, roles and papers behind it.

Detection • Tracking • Re-identification

Florence-2 • Qwen2-VL • Llama 3.2 Vision

Hybrid retrieval • Intent classification • STT / TTS

Sonar • Near-infrared • AUVs & ASVs

Gaussian Processes • XGBoost • LOOCV

03 / Projects

Featured Projects

04 missions

Mission 01↗

+18%

answer relevance

Multi-Channel Conversational AI Platform

LLMs · Hybrid RAG · Voice

Mission 02↗

5

vision tasks benchmarked

Vision-Language Models vs Task-Specific Detectors

VLMs · Computer Vision · Benchmarking

Mission 03↗

2

IEEE papers

Maritime Perception with Sonar & Near-Infrared

Autonomous vessels · Object detection

Mission 04↗

0.996

cross-validated R²

Hybrid ML for a Methane-Coupling Plasma Reactor

M.Tech thesis · ML for Science

04 / Experience

Flight log

  1. Sep 2024 – Jul 2025

    Project Associate

    LFOVIA Lab, IIT Hyderabad

    Evaluated and fine-tuned vision-language models (Florence-2 Base/Large, Llama 3.2 Vision and Qwen2-VL) for multimodal text-image understanding, and benchmarked task-specific models including RetinaFace, DeepFace and Siamese architectures for targeted detection, face analysis and visual tracking / re-identification.

    VLMsFine-tuningFace analysisRe-identification

  2. Nov 2023 – Jul 2024

    Junior Research Assistant

    TIHAN, IIT Hyderabad

    Worked on multimodal maritime perception: developed and curated near-infrared and sonar datasets, and built sensor-specific preprocessing, annotation, data-quality and model-training pipelines for robust object detection in complex maritime environments, resulting in an IEEE conference publication.

    Maritime perceptionNear-infraredSonarObject detection

  3. Jul 2023 – Nov 2023

    Research Intern

    WINET Lab, IIT Hyderabad

    Built foundational sonar-based underwater object-detection datasets with annotation, Python preprocessing, data augmentation and markup pipelines to improve data quality and model robustness. Published at IEEE i-COSTE 2023.

    SonarDataset curationData augmentation

05 / Recognition

Published & awarded

IEEE i-COSTE 2023

Novel Approach to Underwater Object Detection Using Sonar Sensors for Autonomous Underwater Vehicles (AUVs)

Read on IEEE Xplore
Paper

Wanekar A. D., Mannam N. P. B., Rajalakshmi P.

IEEE i-COSTE 2023

Object Detection and Classification for Autonomous Surface Vehicles (ASVs) Through Near-Infrared Imaging

Read on IEEE Xplore
Paper

Wanekar A. D., Mannam N. P. B., Rajalakshmi P.

Hackathon

2nd Place

AI and X Hackathon, IIT Hyderabad

06 / Toolkit

Tools of the trade

Drag to spin

  • Languages: Python, SQL
  • ML: PyTorch, TensorFlow, CUDA, scikit-learn, NumPy, Pandas, Gaussian Processes, XGBoost, Random Forest
  • Vision: YOLO, CNNs, RetinaFace, DeepFace, Siamese networks, Object detection, Tracking, Segmentation
  • VLMs: Florence-2, Qwen2-VL, Llama 3.2 Vision, Multimodal data
  • LLMs: Transformers, LLM fine-tuning, RAG, Hybrid retrieval, Intent classification, STT / TTS

07 / Now · updated October 2026

Exploring how vision-language models and agentic systems can give autonomous systems and edge AI a higher-level, semantic understanding of the world.

  • Completing my M.Tech thesis on ML-guided reactor optimization at IIT Hyderabad
  • Building with agentic AI: systems that perceive, reason and act
  • Looking at how multimodal models can run closer to the edge

08 / Beyond the lab

Curiosity, off the clock

Curiosity does not stop when the lab closes. I build things for fun, learn from people who have gone further than me, and I am deeply curious about Japan, its language and its design.

Creativity

Building for fun

Small robot with a cat face on a 2-inch LCD
A desk companion I built: a 2-inch LCD face, speaker, touch and motion sensors.
Purple 3D-printed cat-shaped case
Its 3D-printed cat case, front.

Curiosity

Learning Japan

Handwritten hiragana practice worksheet
Hiragana Lesson 1, in my own handwriting, with pronunciation notes.
Class screen showing the day's hiragana practice set
Today's practice set: vowels and the K-row, stroke order first.
Japanese class with a hiragana chart on the screen
Japanese class: working through the hiragana chart.
Japanese travel guidebook about India
India through Japanese eyes: the Chikyu no Arukikata guide.
Seiko chronograph on a wrist
Japanese design up close: a Seiko chronograph.

Encouragement

Learning from others

Talk by astronaut Group Captain Shubhanshu Shukla at IIT Hyderabad
Listening to Group Captain Shubhanshu Shukla, Axiom-4 pilot, at IIT Hyderabad.
Bookshelf of non-fiction books
Reading beyond AI: startups, money and the mind.
IIT Hyderabad tower at sunrise
Sunrise at IIT Hyderabad.
Campus buildings at sunset
Evening light on campus.

09 / Contact

Let's talk.

Interested in my work, a collaboration, or just want to talk about multimodal AI? Email is the best way to reach me.

gs25mtech14301@iith.ac.in