Akshay KSoftware Development Engineer, Truminds

Akshay K, ML systems engineer. I build ML systems that stay fastin production.

Software engineer in Bengaluru working on computer vision, MLOps and backend services. I take models from a notebook to a monitored, low-latency service: data pipelines, training, TensorRT optimization, deployment and drift detection.

See my work ↓Get in touch →

proposal 01

modelyolov8s-trt

latency—

objectsscanning.

lower image preprocessing latency on a production CV service: 70%
faster real-time inference API, from 4 s down to 1 s: 4×
less manual annotation through an active learning loop: 50%
CGPA, B.E. Computer Science, Bangalore Institute of Technology: 9.6/10

Experience2023 — now

From object detection prototypes to production ML platforms.

I joined Truminds as an intern working on detection models and stayed on as an engineer. Most of my work sits between the model and the user: making inference fast, data reliable and the whole loop observable.

Software Development Engineer 1
2024 — Present
Truminds Software Systems
Building production ML systems end to end: data pipelines, training, optimized inference, deployment and monitoring.
Key engagements
- Rebuilt the image preprocessing pipeline and cut its latency by 70%.
- Integrated a TensorRT-optimized model, raising inference throughput while holding prediction accuracy.
- Added request batching and ran Locust load tests to find the batch size that maximized throughput under concurrency.
- Benchmarked the model on the full evaluation set to confirm production accuracy and consistency.
- Built and maintained several microservices in the distributed deployment.
PyTorchTensorRTFastAPIAirflowMLflowDVCKafkaDockerKubernetesAzure
Software Development Intern
Mar 2024 — Aug 2024
Truminds Software Systems
- Developed object detection models in Python with TensorFlow and PyTorch.
- Built an automated object recognition and labeling tool that doubled annotation throughput.
- Served models through FastAPI for real-time inference, bringing response time from 4 s to 1 s.
- Shipped Streamlit interfaces so teams could try models quickly and prototype faster.
PyTorchTensorFlowFastAPIStreamlit
Data Science Intern
May 2023 — Jun 2023
Exposys Data Labs
- Compared regression models to predict company profit from data on 50 companies.
- Presented the analysis in Power BI dashboards, shortening the time spent reading the data.
- Delivered the project inside its two-week deadline.
scikit-learnpandasPower BI

How I work6 stages

The whole loop, not just the model.

A model is only useful once it is fed clean data, served quickly and watched in production. This is the lifecycle I have built at Truminds, with the result each stage produced.

Tap a stage for details

Stage 01 of 06Ingest

Streaming ingestion and versioned datasets, so every model can be traced back to the exact data it saw.

KafkaSQLDVC

ProjectsSelected

Things I built outside the day job.

Earlier work across NLP, vision, mobile and web. Most started in a hackathon or a semester and taught me something I still use.

NLP · Full stack2024

Automated Interview Platform

An adaptive interviewer that changes its questions based on what the candidate has already said.

LangChain conversation flows made candidate evaluation 30% faster.
BERT and Word2Vec similarity improved answer scoring accuracy by 40%.
Fuzzy matching handled typos and phrasing, lifting match accuracy by 25%.

PythonLangChainBERTFastAPIReact

Computer vision2023

Vehicle Detection & Counting

Detects, tracks and counts vehicles in traffic video, and estimates their speed.

YOLOv8 detections fed into a lightweight tracker that keeps vehicle IDs stable between frames.
Counts vehicles crossing marked lines and estimates speed from the time between them.

PythonYOLOv8OpenCV

View code

Mobile · Hackathon2023

GenZen

A mental health app with assessments, mood and stress tracking, built by a team of four I led in a 2-day hackathon.

Firebase Auth for secure sign-up and login for 500+ users.
Cloud Firestore backend that scaled with user-generated content.

FlutterDartFirebase

View code

Web2022

Full Stack Blog

A blogging platform with full create, edit and delete flows, shipped by a team of four in two weeks.

Led the team and owned the content management features.

ReactNode.jsExpressMongoDB

Toolkit41 tools

What I reach for.

Python is home. Most days involve PyTorch and TensorRT on one side, FastAPI and the MLOps stack on the other.

ML & vision: PyTorchTensorRTTensorFlowKerasscikit-learnNumPypandasMatplotlib
MLOps: AirflowMLflowDVCCVATDockerKubernetes
Backend & data: FastAPIFlaskNode.jsKafkaCeleryPostgreSQLMongoDBSQL
LLMs: LangChainRAGTransformersOpenAI APIPineconeChromaDB
Languages: PythonC++CJavaScriptTypeScriptDart
Cloud & web: AzureGoogle CloudFirebaseGitReactNext.jsFlutter

BackgroundEducation & more

2020 — 2024

B.E. in Computer Science

Bangalore Institute of Technology

9.6 / 10 CGPA

Community

Core member of the OSCODE club at BIT, running coding contests and hackathons for 100+ students.
Built and launched 15+ responsive websites for college clubs and clients.

Certifications

Machine LearningAndrew Ng, Coursera
Mathematics for Machine LearningImperial College London
100 Days of Code: PythonUdemy
Complete Web Development BootcampUdemy
Flutter Development BootcampUdemy

ContactBengaluru, India

Have a model that needs to ship? Let’s talk.

aksk16902@gmail.com

Akshay K, ML systems engineer. I build ML systems that stay fastp99 ↓in production.