AI/ML Engineer · Generative AI · Multi-Agent Systems · LLM Infrastructure
> initializing agent: aditya_singh
> role: AI/ML Engineer
> status: shipping production GenAI systems
> specialty: making LLMs behave, coordinate, and not hallucinate their way into a bug report
> currently_online: building a real-time voice AI platform
I'm a final-year CS (AI & ML) student who'd rather ship a working agent pipeline than talk about one. Most of my work sits at the intersection of LLM orchestration, RAG systems that don't fall apart in production, and backend infrastructure that can actually handle real traffic.
I care less about "does the demo work" and more about "does this survive a user who types something weird into the input box at 2 AM."
|
A RAG pipeline that questions its users before it trusts them. Prompt injection detection middleware sitting in front of RAG pipelines — catches jailbreaks, role manipulation, and data extraction attempts before they reach the LLM. Ditched a brittle DistilBERT classifier for an embedding + FAISS similarity approach so it adapts to new attacks without retraining.
|
A crypto trading assistant with agents that actually argue with each other. Market Analysis, Sentiment, and Risk agents run in parallel via LangGraph, then a synthesis agent turns their (sometimes conflicting) takes into one trading call. Deterministic services handle the number-crunching so the LLM is only invoked once — because burning tokens to compute an RSI is a crime.
|
Currently building: an outbound voice AI platform where AI agents run parallel phone calls toward a goal you set — sub-600ms latency, real telephony, real barge-in handling. Not a wrapper around an API. The actual plumbing.
Agentic / GenAI
ML / Core
Backend / Infra
Languages
A single well-prompted LLM call is a demo. A system that routes, retries, verifies, and degrades gracefully when the API is down at 3 AM — that's a product.
That's basically my whole engineering philosophy in one line. I'd rather spend a week designing the architecture around an LLM than a day tweaking the prompt.
- 🥇 Top 10 Finalist — Bajaj Finserv National AI Hackathon 2025 (RV College of Engineering, Bengaluru)
- 🥈 Top 5 Finalist — QubitX Hackathon 2025 (GL Bajaj Institute, Mathura)
