Bengaluru, India · AI engineer
Available for freelance engagements
Agents and RAG systems, the self-hosted LLM infrastructure they run on, and the observability that keeps all of it reliable. I work with founders and teams end to end — from first prototype to production, and the support that follows.
Selected work
A selection of systems I've designed and shipped — agents, retrieval pipelines, LLM infrastructure, and the observability that keeps them reliable in production.
An always-on assistant that lives in the chat apps you already use — it books, reminds, researches, and follows up over WhatsApp, Telegram, or Slack. Runs on your own hardware with pluggable models, so your messages and memory stay yours.
An observability layer purpose-built for AI agents — agent-level traces, token-cost views, and failure-mode analysis (loops, tool errors, dead-end runs) on a ClickHouse-backed telemetry pipeline.
A conversational interface over operational databases — LangChain with function calling translates plain-English questions into metrics queries, no hand-written SQL required.
Text-to-video and image-to-video workflows built on ComfyUI — custom node graphs, model management, upscaling, and batch rendering on self-hosted GPUs.
More from the lab
Production retrieval with answers that cite their sources.
LLM root-cause suggestions on live alert streams — MTTR down roughly 25%.
Gemma and Llama-family models via vLLM and llama.cpp, behind a LiteLLM gateway.
CrewAI agents fanning out on parallel research, merging into decision-ready briefs.
On-device question answering that works with no network at all.
Services
End-to-end engagements: I design and build the system, deploy it to your infrastructure, wire up the monitoring, and stay on for support after launch — with tests, documentation, and dashboards included.
Scoped agents that do one job reliably — tool use, memory, evaluation criteria, and guardrails included. From feasibility assessment to a deployed, monitored system.
A private, always-on assistant that lives in your chat apps — tuned to your workflows and extended with custom skills and tools. Self-hosted on your hardware, so your messages, memory, and credentials never leave it.
Open-source models running on your GPUs — LLM serving through vLLM and llama.cpp, media generation through ComfyUI — with routing, deployment, and CI/CD, so your data stays on your infrastructure and your costs stay predictable.
The internal tools that quietly save hours: natural-language SQL over your data, document assistants that cite their sources, and troubleshooting copilots on your alert streams.
Agent-level tracing, token-cost dashboards, and failure-mode analysis for AI systems in production — from instrumentation through to the dashboards.
Not everything I take on is a build — if you need a second opinion on an AI decision, or want your team brought up to speed, say so when you write.
Every system I ship ends the same way —
with a dashboard proving it works.
Knowledge sharing
I document what I learn building AI systems — long-form video, working code in the open, and short-form explainers. All of it public.
Talks and walkthroughs on agents, self-hosted LLMs, and the infrastructure underneath them — recorded as I build.
GitHubThe repositories behind the systems above — agents, MCP servers, and infrastructure experiments, readable end to end.
InstagramClips and quick explainers — the short-form version of the same ideas, for a faster read.
About
I'm Praneeth V P, an AI engineer based in Bengaluru. I've spent my career close to production — large-scale enterprise systems, monitoring, and incident response — and I bring that discipline to every AI system I build.
I run open-source models on my own GPU infrastructure and ship product work out of my home lab — from agents and retrieval systems to personal assistants and video generation pipelines.
Away from the terminal, I'm usually out with a camera looking for birds.
Tell me what you're trying to build. I'll reply with an honest assessment — feasibility, the right approach, and what it would take to ship it.