This page lists projects that I have worked on. Many of these are programming ones and so are code repositories. I recently left GitHub as my primary forge and am trying to build this page as the new landing page.
You should complement these lists with my blog and journal since some projects of mine are just present there as write ups.
1. Statistics
A few overall statistics based on the repositories listed on this page.
1.1. Language Use Statistics
Removing HTML and Jupyter Notebook. It seems like there are a few problems here and I might need to blacklist a few files. For example, I don't think I have written more TeX than Emacs Lisp. Also JS seems to be over-represented.
2. Projects
Here I am nesting projects under some crude first-pass categorization, at the cost of some overlaps.
I run a crawler over a master data file that helps me generate this page. You might find it helpful to use the Table of Content on this page for jumping to different sections of interest.
2.1. Machine Learning
Machine Learning and Artificial Intelligence has been my primary employable
skill. This also means many categories of my work could have some flavor of
ML. While I have tried to put out some broad items here, it's recommend to do a
search for #ML tag on this page to know items outside of this section. Also
check out blog and journal posts with ML tags.
2.1.1. Speech Tech & Conversational AI
Many of the projects in this area are from the time I was working at Skit.ai, 2018-2024. Because of this reason certain descriptions might not make sense when read in isolation. I will improve them with time.
Safety layers for LLM-driven bots at Skit.ai. Includes real-time, in-call, and post-call models, privacy and compliance enforcement, monitoring pipelines for safe conversations.
Migration of Skit.ai voicebots from FSM-based dialog manager to LLM-driven ones. Includes experiments with SFT and post-training, evaluation of different integration strategies for reliable state tracking, production deployment using ICL-based general purpose LLMs.
Core NLP modules for Indian language voicebots: grapheme-to-phoneme converters, entity parsers, rule-based grammars, etc.
Large-Vocabulary ASR systems for 10+ Indian languages (including Indian English). Data collection, training, and deployment at Skit.ai, with production use primarily in Hindi and English.
Finite State Machine based dialog system for live voicebots. Includes ASR, TTS, and NLU models built in-house at Skit.ai. Production-scale deployment provided insights into real-world Conversational AI bottlenecks.
Side models for voicebots: backchannel detection, spoken language ID, noise detection, and runtime integration with production data.
Systems for analyzing human-human and human-machine calls at Skit.ai. Multiple levels of automation from manual inspection to ML-driven tagging.
Research and prototypes for paralinguistic features like non-lexical vocal cue detection and replication, demographic inference, and multi-turn entrainment models.
Evaluation framework and processes for production voicebots. Involves metrics and datasets for component/system evaluation, tooling for rapid experiments and monitoring.
Ink (narrative scripting language for games) based call-flow simulator.
2.2. Research Publications
These are noted in this page.
2.3. Emacs Packages
I use Emacs as my main programming and writing environment. Here are tools and packages that I have written to assist me in different ways.
Org-Roam extensions to support better workflow for links and connections between nodes.
This was also demoed in EmacsConf 2024.

A natural language layer over Matplotlib to visualize data using an LLM.
This was also presented in EmacsConf 2023.
Get live cricket scores and scorecards from cricbuzz in emacs.
Org-Roam extensions to support better workflow for links and connections between nodes.
This was also demoed in EmacsConf 2024.

A natural language layer over Matplotlib to visualize data using an LLM.
This was also presented in EmacsConf 2023.
Get live cricket scores and scorecards from cricbuzz in emacs.
2.4. Epidemiology
While at Reichlab, I worked on time-series modelling for flu and epidemic prediction. Also worked on some visualization tools for the same.
2.5. Visualization
A few information visualization projects and tools.
An exploration of biological interactions based on the homonymous book by Philip Ball, in collaboration with Ipsa.

A natural language layer over Matplotlib to visualize data using an LLM.
This was also presented in EmacsConf 2023.
Code to recreate my library art poster.
2.6. Music Tools
Somehow I have ended up working on a bunch of tools for listening to music throughout the years. Here are they:
Content resolver for mbzlists XSPF playlists.

Music player that lets you explore and play your beets library in acoustic feature space.
2.7. Talks
These are noted on the talks page here.
2.8. Employment Specific
2.8.1. Skit.ai, previously known as Vernacular.ai
If you want an executive summary of my work at Skit.ai, go on my employment page here. I worked at Skit.ai from 2018-2024 as the founding and head of Machine Learning.
My primary technical contribution was around the voicebots that we built. All those projects are noted in the section here. Here I am listing certain broader and non ConvAI contributions that I made at the company. A few small items might be under 2.13.
ML-driven debt-collection workflow at Skit.ai. Includes financial modeling around lead quality, decision systems for timing and method of contact, and integration with main the voicebot stack.
Driving and maintaining organizational culture and core processes at Skit.ai as core part of the leadership team. Includes many initiatives beyond ML or tech.
Employee Satisfaction (ESAT) and other people management processes at Skit.ai.
2.9. Note Taking, News Reading, Planning, etc.
Personal tools for taking notes of various kinds, working with and reading internet news, daily planning via calendars, etc.
Integrating Supernote with my zettlekasten.
2.10. Dotfiles & Ricing
A few older rices are kept here.
Personal variant of Iosevka with larger width and a few non default parameters.
Inner and outer gaps for StumpWM, now part of contrib.
Dark color scheme for Jupyter Notebook based on the emacs theme gotham-theme.
2.11. Websites & SSG
Gimp Script-Fu for exporting favicons as provided by http://www.favicomatic.com/
2.12. Hackathons
Few old hackathon projects from my Bachelor studies (2012-2016).
2.13. Uncategorized
These are items outside of large categories. A few could probably be categorized but I am just dumping all in one group. They make up for good conversations about my interests and non-interests.
JustBooks Catalog as downloadable dataset.
A toy Lisp to python transpiler in the spirit of sxc.
Indoor escape room framework for Home Assistant.
Poor gets the file tree of given github repository and lets you do selective download. Made to learn nim lang.
Utilities for the Hy programming language.
Prototype for Streaming SQL API that allows an Anytime algorithm for statistical queries.
Tool to speed up python script startup time by pre-loading modules in a server.
Attempt at building inexpensive open-design writing instruments and inks.
Small dataset tracking the stench that comes from Karnataka Compost Development Corporation in the nearby area.
My family tree as far as I know.
Mathematical analysis of group selection's effect on robustness in evolution.







