01
Engineering
Building scalable, production-oriented ML and backend systems.
I buildmachine learningsystems.
Currently pursuing an M.Tech in Computer Science Engineering, exploring machine learning through research, experimentation and real-world deployment.
Soumya Chakraborty
The research direction
What am Itrying tosolve?
I am interested in building machine learning systems that do more than achieve a benchmark score. My focus is on understanding the problem, designing useful representations, developing models, evaluating them rigorously, and ultimately making them work in real environments.
[1]The areas below are directions of active exploration rather than claims of established expertise.
Areas of exploration
- 01
Computer Vision
Detection, representation, video understanding
- 02
Deep Learning
Architectures, training dynamics, generative models
- 03
Reinforcement Learning
Policy optimisation, reward design
- 04
Multimodal Learning
Cross-modal representation and alignment
- 05
Time-Series Analytics
Sensor streams, anomaly and event detection
- 06
Machine Learning Systems
Training, serving and system reliability
- 07
Real-Time AI
Latency-bound inference in operational settings
- 08
Applied AI
Models under real-world operating constraints
Currently
M.Tech in Computer Science & Engineering
Building toward a career at the intersection of Machine Learning Engineering and AI research.
02
Research
Investigating new approaches in computer vision, deep learning and intelligent systems.
03
Experimentation
Turning ideas into reproducible experiments, benchmarks and working prototypes.
Industry experience
A model is onlypart of the system.
Ingestion, storage, latency and failure modes decide whether a model is useful — all of it built inside one R&D role.
Hi-Tech System & Services Ltd.
Software Engineering & Machine Learning Intern
- Period
- Deployed for
- Engagements
Eleven months in an R&D division building systems installed on live industrial sites — a steel plant, a utility operator, an ammonia terminal and power boilers — where the cost of a missed detection is measured in safety and production, not validation loss.
Further engagements under this role
- 01RFID Inventory Management5,000+components tracked200+daily users
- 02Incident Response Pipeline40%faster mean-time-to-respond25→15 minresponse interval
- 03Camera Calibration System±0.5°positional accuracy4×improvement over manual baseline
- 04Ammonia Tanker Level Measurement80.77%automatic hard-stop setpointNon-contactno shell penetration
- 05Acoustic Steam Leak DetectionEarlydetection before pressure dropOnlinecontinuous monitoring
- 06Substation Monitoring500+sensor signals / minute10%transformer downtime reduction
Independent projects
Before the model,there isthe system.
Self-directed work, built outside any employer. High-level frameworks hide the cost of every operation — writing what sits underneath them is the fastest way to learn where that cost lives.
The engineering stack
The toolsare notthe point.
Listed as an index rather than a scoreboard — these are the things that have been used to build something real.
Technical index
01
- Python
- C
- SQL
- Bash / Shell
02
- PyTorch
- Scikit-learn
- Computer Vision
- YOLOv10
- R-CNN
- GANs
- Reinforcement Learning
- Transfer Learning
- Model Deployment
03
- Flask
- REST APIs
- Microservices
- PostgreSQL
- MySQL
04
- Pandas
- NumPy
- Stream Processing
- Data Warehousing
- Real-Time Analytics
05
- Google Cloud Platform
- Compute Engine
- Cloud Storage
- BigQuery
- Docker
- Git
- CI/CD
- GitHub Actions
06
- Linux System Administration
- FFT Signal Processing
- RFID Integration
07
- OpenCV
- Matplotlib
- Seaborn
- Jupyter
- VS Code
[1]Ordered by domain, not proficiency. Every entry below has shipped inside a deployed system, a research experiment or a published write-up referenced elsewhere on this site.
About
Why ML?
I am interested in the part of machine learning where theory becomes an actual system — why a model works, where it fails, and what happens when it leaves the notebook.
The next chapteris stillbeing written.
More experiments. More questions. Better systems.