AKSHIT

ModelsAkshit in three-quarter profile, wearing aviator glasses and a dark bomber jacket over a cream knit, against a blurred field of warm orange bokeh.IntoSystems

AvailableWorldwide
  • Deep Learning
  • Edge & CUDA
  • Applied Research

Est2026

I combine research, systems engineering, and low-latency deployment into a single pipeline

The A.I.Y.A avatar — a stylised figure seated on a glowing blue network cube, captioned 'Just another conversational AI'.Latest ProjectA.I.Y.A — An end-to-end multimodal conversational agent: a Python ASR/LLM/TTS voice pipeline wired to a WebGL rendering engine that lip-syncs a live avatar.
Akshit Srivastava, Software Engineer Trainee at Wireless4Scale Lab, IIT Delhi.
The A.I.Y.A avatar — a stylised figure seated on a glowing blue network cube, captioned 'Just another conversational AI'.
The Q.U.O.R mark — three robot faces linked by dotted lines into a triangle around a central checkmark, representing the majority-voting consensus between three language models.
A studio render of an edge compute module — a brushed-metal chassis with a bare accelerator board propped against it, its die exposed at the centre.
Omdena project card for "Monitoring the Water Quality in Bhopal Region Using Satellite Imagery and GIS Techniques", alongside a render of Earth from orbit with a satellite in the foreground.
Omdena project card for "Automated AI-Based Road Inspection System for Detecting and Addressing Road Defects in India", alongside a photograph of a busy Indian street filled with buses and traffic.
IIT DelhiISRO / NRSCVIT BhopalOmdenaDST · VIBHA
About Me

I'm Akshit, an AI engineer focused on models that hold up outside the notebook.

Right now that means deep learning for 5G fault analysis at the Wireless4Scale Lab, IIT Delhi — pipelines running on NVIDIA Jetson AGX modules, with the hot paths rewritten as custom CUDA kernels. Before that, ML/DL early-warning systems over geospatial meteorological data at ISRO.

Social +

Est2026

Selected Works

2024—2026

The Q.U.O.R mark — three robot faces linked by dotted lines into a triangle around a central checkmark, representing the majority-voting consensus between three language models.02

Q.U.O.R

Research build

Quorum for Unified Opinion and Resolution

A multi-agent LLM council where three models reason independently, challenge each other, and converge through majority voting.

3
Model council
FAISS
Vector search
  • Python
  • PyTorch
  • LangChain
  • Sentence Transformers
  • Groq API
  • Hugging Face Transformers
Omdena project card for "Automated AI-Based Road Inspection System for Detecting and Addressing Road Defects in India", alongside a photograph of a busy Indian street filled with buses and traffic.03

AI Based Road Inspection System

Open source

Road anomaly detection at street scale

Omdena VIT Bhopal × Omdena Mexico Chapter

Fine-tuned object detection for potholes, cracks and alligator-cracks, shipped as an interactive dashboard.

32.6%
mAP@0.5
  • Python
  • YOLOv8
  • Streamlit
  • Roboflow
  • OpenCV
Omdena project card for "Monitoring the Water Quality in Bhopal Region Using Satellite Imagery and GIS Techniques", alongside a render of Earth from orbit with a satellite in the foreground.04

AI Based Water Quality Monitoring

Open source

Satellite-driven water quality analysis, Bhopal region

Omdena VIT Bhopal Chapter

A satellite imagery pipeline and analytics dashboard for monitoring water quality across the Bhopal region.

  • Google Earth Engine
  • Data Analysis
  • Data Management
  • Tableau
  • Streamlit
  • 0
    Publications
    ACRS 2026 · Research Square
  • 0
    Technical Articles
    Published on Medium
  • 0
    Engineered Projects
    Agents, vision, geospatial
  • 0
    Research Labs
    IIT Delhi · ISRO

Akshit

Right now that means deep learning for 5G fault analysis at the Wireless4Scale Lab, IIT Delhi — pipelines running on NVIDIA Jetson AGX modules, with the hot paths rewritten as custom CUDA kernels. Before that, ML/DL early-warning systems over geospatial meteorological data at ISRO.

Akshit is really good at AI/ML, especially computer vision. He helped me a lot during a project by suggesting the right approach, fixing issues, and improving the overall results. Super easy to work with and always ready to help.
Akshad AgarwalTechnical Consultant @ EY

Peer-reviewed output 2 papers

01

A fully-connected neural network diagram: an input vector x feeding three hidden layers of nodes joined by weight matrices W1 to W4, resolving into an output vector y.

Deep Learning Systems

  • PyTorch
  • JAX
  • TensorFlow

Architecture design and training for real-time monitoring and fault analysis, built against a latency budget rather than a leaderboard.

02

An isometric diagram of edge deployment: a cloud and data-centre racks on the left, a control-plane pane in the middle, and a scatter of green edge servers in remote locations on the right.

Edge & Accelerated Compute

  • CUDA
  • Jetson AGX
  • Edge ML

Getting models onto the hardware that actually ships — Jetson AGX-class edge modules, with critical components hand-written as CUDA kernels.

03

A hand-drawn diagram titled 'Language + Agentic AI': natural language and speech feeding a robot at a laptop, which fans out into a plan, act, observe, adapt and achieve loop, captioned 'from understanding to outcome'.

Language & Agents

  • LangChain
  • RAG
  • Multi-agent

Multi-agent LLM councils, RAG retrieval over vector search, and multimodal ASR/LLM/TTS pipelines that hold a live conversation.

Akshit buildingmodelsA plate of dark rippled fabric lit so the folds read as silver ridges, captioned "Language — understanding, reasoning, communicating" and "Agentic AI — planning, acting, adapting, achieving".thatsurvive contactwith the real world

A model isn't finished when it converges, it's finished when it runs inside the latency budget, on the hardware you actually have, in front of a real user.

ResearchInterests2026

01Natural language pragmatics and language driven control in robotics
02Computational models and architecture for language understanding in LLMs
03Embodied Human-AI Interaction
04Deep learning architecture design
05Applied ML for real world problems
06Mechanistic Interpretability

From research to deployment, every step is measured

01

Frame

Read the literature, define what "working" means numerically, and pin the constraint that will decide the design — usually latency or memory.

02

Build

Train, profile, and cut. Most of the work is finding which part of the model is actually costing the budget, then rewriting that part.

03

Ship

Deploy to the target hardware, measure on device, and write it up — as a dashboard, a preprint, or a paper.

Positions

RRSC North, NRSC, ISROCompleted

Research Intern

Oct 2024 — Dec 2024

  • Designed and integrated an ML/DL based pipeline for early warning systems over meteorological data to monitor vector-borne disease outbreaks, reducing manual analysis time by automating feature extraction across various geospatial datasets.
  • Performed multivariate time-series analysis to quantify seasonal and environmental correlation for disease incidence.

Resulting paper

A. Srivastava, K. Mirza, S. Saran — 47th Asian Conference on Remote Sensing (ACRS), 2026. Abstract submitted.

  • Python
  • Scikit-learn
  • TensorFlow
  • QGIS
  • ArcGIS
  • Pandas
  • NumPy
W4S Lab, IIT DelhiCurrent

Software Engineer Trainee

March 2026 — Present

  • Developed deep learning algorithms for a real-time monitoring and fault analysis system for 5G wireless networks in an R&D environment.
  • Deployed the algorithms and Machine Learning (ML) pipelines on NVIDIA Jetson AGX-class edge computing modules to target low-latency telecom applications.
  • Implemented critical components of the algorithm using custom CUDA kernels to accelerate complex computations and ML workloads.

Hot path

Critical components hand-written as custom CUDA kernels to accelerate complex computations and ML workloads.

  • Deep Learning
  • CUDA
  • NVIDIA Jetson AGX
  • 5G Networks
  • Edge ML

Research & Writing

2026Abstract submitted

Effect of Meteorological Factors and Seasonal Dynamics on Dengue Incidence in Haryana Region

A. Srivastava, K. Mirza, S. Saran · 47th Asian Conference on Remote Sensing (ACRS)

Quantifies how seasonal and environmental variables correlate with dengue incidence across the Haryana region, using multivariate time-series analysis over geospatial meteorological data.

Technical Writing

Toolkit

Languages

6 listed

  • Python
  • C++
  • C
  • SQL
  • JavaScript
  • CUDA(Beginner)

ML / DL Frameworks

7 listed

  • JAX
  • PyTorch
  • TensorFlow
  • Keras
  • Hugging Face Transformers
  • scikit-learn
  • LangChain

Tools & Infrastructure

5 listed

  • Git
  • Docker
  • FastAPI
  • Streamlit
  • Plotly

Data & Visualization

5 listed

  • Tableau
  • Power BI
  • MongoDB
  • ArcGIS
  • QGIS

Education

B.Tech. Computer Science Engineering (AI & ML)

VIT Bhopal University

8.38/10CGPA · 2026

Leadership & Service

  • General Secretary

    Data Science Club, VIT Bhopal · 2025 — 2026

  • Technical Team Member

    Data Science Club, VIT Bhopal · 2024 — 2025

  • Student Coordinator, Project Exhibition

    Industrial Conclave, VIT Bhopal · Jul 31 — Aug 03, 2024

Honours

Runner-up, Space India Hackathon, IISF’23

Department of Science & Technology (DST), Government of India, and Vijnana Bharati (VIBHA)

Jan 2024

Testimonials

Word of Mouth

I’ve known Akshit since our college days, where I was his senior by one years, and I’ve consistently been impressed by his problem-solving ability. He has a strong aptitude for breaking down complex problems into smaller, manageable components and solving them logically and efficiently. I was able to clearly observe this while working closely with him, his approach is structured, calm, and solution-driven rather than trial-and-error. Akshit picks up new concepts quickly and applies them with clarity, which makes him dependable in high-pressure or ambiguous problem statements. His analytical thinking, combined with execution speed, sets him apart from many peers at a similar stage. I would strongly recommend Akshit to any team looking for someone who can think clearly, adapt fast, and deliver results.
AKAnniruddha KumarData Scientist @ Flodata
Akshit is really good at AI/ML, especially computer vision. He helped me a lot during a project by suggesting the right approach, fixing issues, and improving the overall results. Super easy to work with and always ready to help.
AAAkshad AgarwalTechnical Consultant @ EY
We reconnected only recently, though we were schoolmates, and since then he has been a consistent source of support and guidance. He is welcoming, warm, and always willing to help, often going out of his way to do so. What stands out most is his ability to offer thoughtful assistance even beyond his own field of expertise, reflecting both intellectual openness and genuine concern for others’ growth. He is passionate about his work, highly hardworking, and remains remarkably humble in his approach. His sincerity, reliability, and quiet dedication make him someone one can truly depend upon.
RSRiya SinghStudent @ Christ University

Want to contact me?

Currently

Software Engineer Trainee at Wireless4Scale Lab, IIT Delhi, open to research and engineering conversations.

Get in touchDownload CV