Software Engineer · Bengaluru, India
I design and build dependable backend systems and applied AI products — the quiet infrastructure that lets good products stay good under pressure.
I'm a backend-leaning full-stack engineer. My work lives mostly in the parts of a system no one sees in a demo — APIs, data models, retrieval pipelines — yet decide whether it holds up.
I've written production code at an early-stage startup, delivered client work as an independent developer, and built things on my own initiative end to end: a real estate platform, an npm package that makes API latency legible, a fine-tuned language model, and a retrieval system grounded in regulatory text.
Lately my interest is in applied AI that has to be correct, not just impressive — grounding LLM output in real data, measuring what models actually do, and shipping the result as ordinary, maintainable software.
A fine-tuned DistilBERT model predicting a food product's Nutri-Score grade (A–E) from ingredient text alone. Labels were computed from the official formula rather than trusted from source data, trained on ~926k global products, and evaluated on a held-out India set to surface a genuine domain-shift gap.
Python · PyTorch · DistilBERT · Hugging Face
A retrieval-augmented question-answering system over Indian food-safety regulations. Plain-language questions return answers grounded in the actual regulatory text, using hybrid search (BM25 + vector embeddings) so exact clause matches aren't lost to pure semantic retrieval.
Python · OpenSearch · Gemini API · Hybrid Search
A published npm package that decomposes request latency into its components — database time, server compute, network overhead — so performance work starts from measurement instead of guesswork.
JavaScript · Node.js · npm
A full-stack property listing platform with search and multi-criteria filtering, built solo end to end: data modelling, REST API, authentication, and frontend.
MongoDB · Express · React · Node.js
An independent engagement delivering a listing website for a client — requirements gathered directly, shipped iteratively, and maintained in production. Code is client-owned.
Smaller experiments — a web crawler, a bird-flock simulation, a weather forecaster — live on GitHub.
Delivering client work and personal systems as an independent developer — owning requirements, architecture, and delivery directly, with ongoing maintenance and new features in the pipeline.
Worked across the stack on a production product, leaning backend — APIs, database design, and the services beneath the parts users see.
Node.js · Express · FastAPI · Spring Boot · Go (working knowledge)
MongoDB · PostgreSQL · OpenSearch · Vector search
RAG pipelines · Embeddings · Hybrid search · Model fine-tuning
React · Next.js · JavaScript · HTML / CSS
Git · Docker · Linux · npm publishing · REST API design
Few things, done carefully.
Whether you're building a product, backing one, or simply want to talk systems and technology — my inbox is open.
vishsecures@gmail.com Resume — PDF