Python Developer — AI Engineer
Sindhukumari Parameswaran
I build systems that turn signals into understanding — from spiking neural networks to on-device translation.
>learning Gen AI, RAG, Agentic AI, MCP
I build systems that turn signals into understanding — from spiking neural networks to on-device translation.
5+ years of experience across the full software development lifecycle — requirements gathering, design, development, implementation, and maintenance — using Python and Java. Strong hands-on experience with Django and Flask, RESTful APIs, SQL and relational databases (MySQL, Postgres), and AWS (EC2, S3). I've built and maintained CI/CD pipelines end to end, integrating Git to enforce code quality and automate releases.
Right now I'm moving deeper into Gen AI, RAG, Agent AI, Agentic AI, and MCP, alongside ML, DL, and data engineering — extending the same AI/ML foundation from NeuroCAD and the other projects below into where the field is heading next.
// at the desk, mid-build
Python Developer
Neurolabs Innovations Pvt Ltd — Project TRAI
Engineer (R&D)
ORBAI
Software Developer
FuGenx Technologies Pvt Ltd — USM AI Recruitment Platform
Every project below is a pulse on the same line — some shipped to production, one still open-source and growing.
An on-device overlay that translates chats in WhatsApp, Snapchat, and Messenger without sending a word to a server. The core runs on Python and Argos Translate, wrapped in platform-specific shells — Kotlin and Chaquopy for Android, keyboard and share extensions for iOS, PyQt6 with OCR for desktop, and a Chrome extension for the web. The monorepo, translation core, CI, and Android scaffold are done; right now I'm testing a two-stage pipeline that runs romanized Hindi through IndicXlit before handing it off to Argos Translate for Hinglish support.
Built at ORBAI — a GUI tool simulating human brain processes with spiking neural networks, using 2D Gaussian filtering for images and Perlin noise for video. Presented at NVIDIA's GTC conference.
Built with a team of 3 to predict legal outcomes and auto-generate filing documents, using Law Stats, NLP, and Legalese frameworks, deployed on AWS.
Automated resume parsing with Stanford NER and a Gensim-based candidate-ranking model, plus a recruiter chatbot on Google Dialogflow.
A Jaccard-similarity engine that scores incoming MAC addresses against a trusted database to flag spoofing attempts.
Presented at NVIDIA's GTC Conference, March 2019 — NeuroCAD, a design tool for spiking neural networks.
The throughline in my work is turning a signal into something that acts. Here's where I want to take that next.
NeuroCAD is where this started — designing spiking neural networks and presenting the tool at NVIDIA's GTC. It's still the base I build everything else on.
Universal Robots · CC BY-SA
NeuroCAD was about designing spiking networks on screen. I want to take that same neuromorphic approach into real-time robotic control — low-power, low-latency decisions running on the machine itself.
NASA · Public DomainLive Chat Translator runs entirely on-device, by design — no server round-trip. That same constraint, autonomy without constant connectivity, is exactly what onboard AI for space systems needs.
NeuroCAD is where this started — designing spiking neural networks and presenting the tool at NVIDIA's GTC. It's still the base I build everything else on.
Universal Robots · CC BY-SA
NeuroCAD was about designing spiking networks on screen. I want to take that same neuromorphic approach into real-time robotic control — low-power, low-latency decisions running on the machine itself.
NASA · Public DomainLive Chat Translator runs entirely on-device, by design — no server round-trip. That same constraint, autonomy without constant connectivity, is exactly what onboard AI for space systems needs.
Open to conversations about Python, AI systems, or the kind of on-device work Live Chat Translator is built on. The fastest way to reach me is email.