A two-hour workshop that turns fragmented threat feeds into an analyst-ready, agent-assisted investigation workflow.
- cybersecurity
- databricks
- agents
- threat-intelligence
Hi, I’m Praneeth.
I build production AI systems, evaluate how they behave, and write about the engineering work between a promising demo and a dependable product.
I work as a Solutions Architect focused on GenAI and agent systems at Databricks. My recent work covers agent evaluation, autonomous AI engineering, document intelligence, real-time agents, and technical workshops.
This site is where I keep the implementation details: what I built, what failed, what the measurements changed, and what another engineer can reuse.
Latest writing
All essays and notesA visual, systems-level breakdown of Gemma 4 architecture choices and what they suggest about open model design.
- gemma
- llm-architecture
- open-models
- multimodal
Why enterprises need a control layer above individual coding and agent runtimes.
- agents
- harnesses
- omnigent
- enterprise
A small trajectory rule that caught a large class of polished agent failures.
- evaluation
- agents
- traces
- citations
A practical map of what changes when an agent leaves the notebook and starts serving real users.
- agents
- evaluation
- mlflow
- production
Selected projects
All projectsCLEAR-S Eval Harness
A layered evaluation system for testing agent outputs, trajectories, safety, latency, and cost across prompts, frameworks, and models.
- MLflow
- agents
- evaluation
- traces
Databricks AI Intern
An autonomous AI and ML engineer that researches, trains, measures, reproduces, and serves what it builds using Databricks-native primitives.
- Databricks
- MLflow
- agents
- serverless
LayoutScribe
A Python package that converts PDF, PowerPoint, and Word files into Markdown and layout-aware JSON using multimodal models.
- Python
- multimodal
- documents
- PyPI
Real-Time Agent APIs
A reference implementation for streaming agent events, tool activity, approvals, and errors through FastAPI and WebSockets.
- FastAPI
- WebSockets
- Python
- agents
Workshops and teaching
Hands-on material for teams moving from AI concepts to working systems.
A two-hour workshop that turns fragmented threat feeds into an analyst-ready, agent-assisted investigation workflow.
- cybersecurity
- databricks
- agents
- threat-intelligence
Workshop notes for moving from request-response agents to observable, event-driven systems.
- agents
- fastapi
- websockets
- python
I prefer measured claims, practical code, and explanations that keep the tradeoffs visible.