I am a research-focused engineer who explores capability of present-gen AI models and architectures. I love spending quality time with the people I love. I try to build things my 80-year-old self would be proud of.
-
Building knowledge graphs of people from public information. I think there is a lot of room here. Some of this work is closed-source, but I would love to collaborate.
-
Working on LLM inference layers. I am especially interested in how RAM and VRAM can be used together while keeping the GPU focused on the main model computation.
-
Thinking a lot about whether neural networks can get closer to actual reasoning. One direction I keep coming back to is
IDEA-AS-A-TOKEN. -
Reading about neuromorphic computing and using some of those ideas as reference points for the work above.
-
👨💻 A few projects I am building:
- ModelServer - Docker-first LLM inference stack for CUDA and ROCm. Just plug and play.
- Graphrag-OSS - GraphRAG using ClickHouseDB and open-source LLM and embedding models.
- Project-Pod - Developing a knowledge graph of a person's professional information from public data. I would welcome collaborators in the future. It is currently in the
brainstorming/initphase.
-
👨💻 Grad school work:
- Low light image enhancement with denoising — image restoration for tough lighting conditions
- Human Robot Interaction — interaction experiments and LLM-assisted robot behaviors
- Voice authentication on Boston Dynamics Spot — applied speech and identity workflows on a real robot
- Evaluation of Kalman-Filter based SLAM Systems - A quantitative analysis of the system-latency vs. ATE tradeoff in vision SLAM systems. This report does not incorporate recent deep-learning-based systems.
-
🌐 Portfolio / projects: sougato97.github.io
-
📫 Let's connect: sougato97@live.com (preferred) / sougato97@gmail.com
-
📄 Resume
- At a B2B US pharma-focused health-tech startup, I served as lead engineer and reported directly to the CEO. My work included building proofs of concept, delivering technical training, and leading research for system design and development.
- I architected a multi-agent system that processed more than 100 GB of US pharma data. I designed its routing layer around control signals, concurrency, and clear separation of concerns.
- At Applied Research Works | Cozeva, a B2B health-tech company, I worked as a Machine Learning Engineer and reported directly to the CTO. I architected AI assistants for enterprise clients that grounded responses in real data and returned source-backed answers.
- I also designed a medical-coding agent that extracted ICD-10 codes from doctors' notes, significantly reducing manual coding time.
- Earlier, as an intern, I worked with Social Determinants of Health data and handled large-scale health datasets under data-privacy laws such as HIPAA.
