Agents that act on infrastructure
Autonomous task execution with AI decision-making: monitoring and alerting, predictive analytics, and natural-language commands that spin up containers, tune database queries and scale resources.
AI Development
I build AI that does work: agents that read data, make decisions and act on production systems, plus the LLM pipelines that feed them. Not chatbots. The goal is a system that keeps running, and keeps getting better, without someone watching it.
01
What I build
Research and analysis pipelines
Scraping from multiple sources, NLP content understanding, LLM summarization, trend detection and personalized briefings, exposed through an API and a web interface.
Domain-specific AI assistants
An AI chatbot built to understand a specific domain. For a trading client who asked for a bot, that meant an assistant that understands markets, on a custom server.
Adaptive systems
Learning algorithms and feedback loops so the system improves with use: users mark outputs useful or not, and that signal drives retraining.
02
Results from shipped work
Sanskriti Labs
Lead Developer & AI Architect · 6 months, Jan 2024 to Jun 2024
- 50% reduction in manual DevOps tasks
- 99.5% system uptime through predictive maintenance
- 40% improvement in deployment speed
Research Agent
Full-Stack Developer & Data Scientist · 4 months, Aug 2023 to Nov 2023
- Processes 500+ papers and articles daily
- 90% accuracy in content classification
- 60% less research time for users; insights led to 3 new research directions
03
What I design for
These are the problems that came up on real projects, and the ones I plan for before writing code.
Reliable decisions in production
An agent that acts must fail safely. Error handling and recovery paths are designed first, not bolted on.
Security versus automation
More autonomy means more blast radius. I scope what an agent may touch and keep the flexibility without handing it the keys.
Real-time performance
Decisions on live systems have latency budgets, so processing is built for the real-time path.
Data quality and source limits
Rate limits and anti-bot measures, summaries that stay good across very different content, and data that stays fresh and relevant.
04
Tools I use
- Python
- FastAPI
- Flask
- TensorFlow
- OpenAI GPT
- spaCy
- NLTK
- Scrapy
- BeautifulSoup
- PostgreSQL
- MongoDB
- Redis
- Celery
- Docker
- Kubernetes
- React
05
Is it a fit?
The best fit is an AI agent that touches production infrastructure, or an LLM pipeline that has to run unattended on a schedule and produce something people actually use.
If you need a quick chatbot widget or a one-off script with no system around it, I'm probably not the right person, and I'll tell you so.
06