|
wd-security-unlocker Linux desktop utility to unlock and mount compatible Western Digital My Passport drives. |
|
docuwise-2 Turns PDFs into a searchable knowledge base with question answering. |
|
forced-move Strategy game where each move changes options for both players. 🎮 Play - Forced Move |
|
venuemeter-poc Azure-based prototype for venue sustainability metrics and non-functional web app testing. |
- I'm a software engineer who loves building to entertain, teach or for convenience. If my software encouraged or enabled concentrated thought in my users, that's an added bonus.
- Examples of areas/domains where I have been able to contribute in the past as a machine learning engineer are :
- Recommendation systems pertaining to luxury and fast fashion products, food and entertainment. Ever optimized your recommendations for customer loyalty and gained insights into how loyalty can be measured? Or used an unique metric as a similarity measure? I'm still very interested in understanding the sense of taste from a neuroscience angle and once took a shot at imitating it in a recipe recommendation system.
- Sleep science detecting and predicting bodily conditions based on biosignals recorded while a person sleeps. Learned to appreciate the healing powers of a good night's sleep through this. Found it funny when I once had a date ask me what my sleep score was.
- Price modelling tasks such as adjusting pricing/developing pricing models for marketplace based businesses Eg: Uber (service oriented marketplace), Amazon (Ecommerce oriented marketplace), Rover (Specialty marketplace). The details where profits hide.
- Natural language processing tasks such as : entity linking, relation extraction, semantic evaluation pertaining to fairy tales and news articles. Ever tried analyzing personas of fairy tale characters ? - it generalizes human behaviour pretty well. News articles are a way to assimilate a lot of signals and can spark action - I particularly worked with enabling traders to make investing/buy/sell decisions by serving events in articles in a way that makes these signals very clear and easy to spot.
- Building intelligent, adaptive systems - These are now agentic systems or workflows. But the desire to automate is really old and there are several other ways an attempt to create adaptive systems can be made. If you ever take part in this pursuit,remember that the path towards this isn't necessarily imitating human form or level of intelligence.
- Retrieval Augmented Generation : ingestion optimization using knowledge graphs and metadata, retrieval and generation optimization techniques including small to big retrieval and others highlighted in the llamaindex framework, decision making around embeddings and appropriate distance metrics, extending RAGs to RAG agentic abilities. A popular investing firm and a couple of hedge funds once paid us a lot of money to build them a magical tool that could sort through an extremely large document base and enable them to do their due diligence and better portfolio management/optimization. It had its advantages but this definitely wasn't a replacement of someone who is able to connect various disjointed ideas and stare into a blank space and suddenly knows what to do. Mastery cannot be poorly imitated.
- Text based Extraction using compound systems: What's the sota score you can achieve in OCR? Does that model end up doing well in your specific business use case? I present to you a system that automatically adapts to your business case and will always give you better results than any sota model you use for OCR.
- Robotics and Embedded Systems - My first love, for a time when I give into the physical constraints of my body.
- My work centers around building robust software systems with end-to-end ownership, embedding machine learning from research through production when it meaningfully advances the product. With a research background in classical machine learning, deep learning, natural language processing, computer vision, topology, and computational neuroscience, I now operate primarily as a full-stack software and AI engineer, delivering secure, scalable platforms with strong attention to usability and real-world effectiveness. More importantly, I believe that the field of software development is ever-evolving and take particular satisfaction in continuously learning, refining and expanding my craft.
- You can reach me at: titas.das+gh@gmail.com
-
Fun fact: Once upon a time in grad school, cooking felt mundane enough that I built a robot to chop vegetables (no fancy cuts). Years later, I've come to find both cooking and cleaning oddly therapeutic. I've grown to appreciate everyday routines as interconnected systems that shape health and quality of life.
-
I’ve grown increasingly interested in everyday environmental health, spending time understanding how dust and air pollution affect indoor spaces and how exposure can be reduced in practice. At the same time, I’ve been learning about the medicinal and biochemical properties of ingredients and bringing that knowledge into my cooking in a thoughtful, intentional way.
-
The images below come from an ongoing exploration of how multimodal LLMs interpret and represent visual features, paired with a more personal motivation: carving out time to paint, an activity I find both grounding and enjoyable.
