Jegan Murali · AI Systems Engineer. I build agentic AI, RAG and voice AI at Ecom Ignite, and take on select contract builds.
 |
AI that takes real actions, with scoped tools, approval paths, evals and monitoring, so it does the job without surprising you. |
 |
RAG and business-data products that return controlled, explainable answers from your own data, not confident guesses. |
 |
Speech-to-speech experiences for education, accessibility and customer operations that feel natural to talk to. |
|
Ask company data anything in plain language. Read-only tools, SQL validation and human approval keep answers safe, and Mira Monitor tracks latency, errors and cost. |
Runs real commands on your machine, so it ships with real guardrails: a blocklist, confirmation before anything disruptive, tests and CI. Groq or fully local with Ollama. View the code → |
|
A speech-to-speech learning assistant for blind children, built on Whisper, LLaMA 70B and fine-tuned TTS. |
CNN + DNABERT-2 (LoRA) + KAN for CRISPR-Cas9 off-target prediction in cystic fibrosis, with an LLM safety-audit summary. Read more → |
Also built: AURA, an AI companion that guides people through emotions, mindfulness and inner-child work before therapy begins.
1. Intro → You describe the problem. I say honestly if AI is the right tool for it.
2. Thin slice → One real task, end to end, fast. You see it work before we commit to more.
3. Build & prove → Guardrails, evals and tests go in before the system grows.
4. Ship & watch → Release with monitoring, then improve from what it actually does.
Systems: agent design · tool orchestration · guardrails · evaluation · observability · RAG
Models: LLaMA · Hugging Face · Whisper · LoRA · TTS · DNABERT-2 · KAN
Every commit raises a skyscraper. Drive through the live 3D version
