AI engineer. I put retrieval, agents and voice models into production.
Now: building an NL-SQL engine over 20+ databases at Atlasml.ai.
This is a map of my own work. Each point is a project, post, skill or role, or a passage inside one, placed so that similar things sit close together. Drag it, hover a point, or pick a topic.
Six projects, from production systems to hackathon wins. The first is the one I would start with.
Retrieval & data2026 - present, at Atlasml.ai
An enterprise multi-agent platform that turns natural language into validated SQL across 20+ heterogeneous databases, with self-healing queries and secure multi-tenant isolation.
FastAPI, CrewAI, gRPC, Docker, PostgreSQL, MySQL, MSSQL, Snowflake
An open-source animated icon library with 1200+ ready-to-use icons for React, Vite and Next.js, built on Framer Motion and Tailwind CSS.
An interactive cybersecurity learning platform with live command execution, attack simulations and a real-time encryption demo.
An AI-powered fashion e-commerce platform with smart recommendations and personalized styling, which took 1st place at a hackathon.
A competitive programming platform with a custom in-browser code editor where students solve problems and collaborate on challenges.
A student platform for skill certification and peer connection over chat and video calls, built at a national-level hackathon and placed 1st.
An honest split of where I am deep and where I am still learning, after two years of building AI systems.
Where I have shipped production systems and know the failure modes.
Hybrid search & reranking; Pinecone, pgvector, FAISS; Chunking strategy & RAG evals; Citation-grounded answers.
OpenAI, Claude, open-weights; Function calling & tool use; LangChain, LangGraph workflows; Streaming, guardrails, cost tuning.
ASR / TTS pipelines (Whisper+); Audio processing for voice models; Real-time streaming audio; VAD & speaker diarization.
FastAPI, Node services; Next.js, React frontends; Postgres + pgvector; Docker, AWS deploys.
Real project work, but less miles. I am still building depth here.
LoRA, QLoRA, PEFT; Dataset curation & cleaning; Instruction tuning; Evaluation loops.
Detection & classification; OCR & document AI; OpenCV pipelines; Image preprocessing.
Four roles since late 2023, drawn to scale. The teaching job ran alongside the first engineering one.
Retrieval & dataAtlasml.aiAI/ML Engineer, April 2026 - Present
Models & agentsCybercom CreationAI/ML Engineer, Jan 2026 - March 2026
Web & toolsSampaarsh TechnologiesSoftware Engineer, August 2024 - October 2025
Web & toolsDarshan UniversityTeaching Assistant (during undergraduation), December 2023 - September 2025
Notes on how AI systems work and why they break, written while I build them.
Ever talked to a bot that paused for two seconds before every answer? That pause is not a bug in the corner, it is the whole experience. Learn how a voice agent hears, thinks and speaks, where the time goes, and why latency is a feature you design for, without the confusing AI jargon.
12 min read
Models & agents
Transformers keep a growing KV-cache “notebook,” while Mamba compresses history into a fixed-size state. This article explains the tradeoff, how Mamba evolved from Mamba-1 to Mamba 3, and why hybrid architectures may be the practical path for long-context LLMs.
5 min read
Retrieval & data
Ever wondered why a model that writes perfect SQL in a demo still ca not be trusted against a real database? Learn what a Text-to-SQL harness actually is. Routing, schema retrieval, validation gates, repair loops and read-only execution - explained with real examples and without the confusing AI jargon.
12 min read
I am an AI engineer with two years of experience building systems that make it to production, not just notebooks. The unglamorous parts are where I spend my time: retrieval quality, latency budgets, evals and clean data.
Day to day that is production RAG, LLM integrations and agent workflows, and voice and audio pipelines, with fine-tuning and computer vision alongside. Lately: a multi-tenant RAG platform, an NL-SQL engine that queries 20+ databases, and a LoRA-fine-tuned vision-language model watching real warehouses.
Away from work I won first place at two hackathons, and I have taught more than 500 students as a teaching assistant at Darshan University. I like leading technical projects and bridging AI research with applications people actually use. I am based in Ahmedabad, India.
I am open to AI engineering roles and to projects. Email is the fastest way to reach me, and I read everything.