Building reliable AI systems and thoughtful software.
I build reliable AI systems, from intelligent agents and retrieval pipelines to production applications.
I am an AI systems engineer based in Bengaluru specializing in multi-agent architectures, agentic retrieval systems, and end-to-end production engineering. I design systems with verified handoffs, deterministic boundaries, and human-in-the-loop governance.
From distributed agent orchestration and vector search to production APIs and high-performance WebGL interfaces, I own systems from architecture to deployment.
B.Tech, Computer Science · 2021–2025 · 78%
Sri Chaitanya Junior College
Intermediate, MPC · 2019–2021 · 94.8%
Autonomous Voice AI Screening & Talent Pipeline Orchestration.
Associate Software Engineer
AI & Backend · Bengaluru
See it built: SMATE APISMATE API↗ SMATE interfaceSMATE interface↗
Backend Intern · Remote
Worked on API performance and training-data preparation.
See it built: Healthcare Records APIHealthcare Records API↗ Credits APICredits API↗ Blog APIBlog API↗
LangGraph · LangChain · AgentScope
Supervisor & routing patterns
RAG · Hybrid retrieval · RAGAS
MCP · OpenAI & Claude APIs
Python · FastAPI · Node.js
React · Next.js · TypeScript
PostgreSQL · pgvector · Redis
Docker · Pytest · Git
Claude Code · OpenAI Codex
Gemini CLI
Typed contracts · Evaluation
Human approval boundaries
A selection of AI experiments, backend applications, and interactive 3D studies.
An end-to-end platform for aligning language models with human feedback: preference annotation, SFT with QLoRA, a Bradley-Terry reward model, and Direct Preference Optimization, served through FastAPI, Celery, and Redis.
An AI agent orchestration terminal for building text-processing pipelines out of specialised agents (sanitizer, briefer, analyst), with multi-LLM fallback across Gemini and Hugging Face, and a Vitest/Playwright test suite.
A speech pipeline that separates speakers (VAD, ECAPA-TDNN embeddings, clustering) and transcribes them across languages with Whisper, confidence scoring, and a human-review API.
An OCR pipeline (PaddleOCR + Tesseract) that uses LayoutLMv3 and a DETR-based table detector to pull structured regions, key-value pairs, and tables out of PDFs and images as JSON.
Ingests metrics over Kafka or Prometheus scraping and runs three anomaly-detection algorithms in parallel — Z-score, Isolation Forest, and an LSTM autoencoder — alerting to Slack and PagerDuty with Grafana dashboards.
A benchmark harness for LLM performance on Indic languages, with adapters for OpenAI, Anthropic, HuggingFace, and Ollama, plus exact-match, ROUGE-L, BERTScore, LLM-as-judge, and hallucination-detection metrics.
An explainable, upload-first AI hiring intelligence platform using bring-your-own-key OpenAI analysis.
A full-stack financial advisor where demo personas track expenses, manage investments, and get AI-driven coaching through OpenAI and Perplexity.
Coordinates team projects with role-based access control (Admin/Member), task tracking, dashboards, and activity logs.
A self-hosted experiment-tracking platform with a Python SDK, FastAPI/gRPC metrics ingestion, and Celery-based drift detection — a lightweight alternative to MLflow or Weights & Biases.
A GraphQL API (Apollo Server, Prisma, Auth0) powering SMATE, the EdTech tutoring concept behind the Smart Tutor proof of concept I built at Terralogic.
The Next.js interface for SMATE, authenticated through Auth0 and driven by the GraphQL API above.
An AI career counselor users talk through career questions with, with session management and a tRPC/Prisma backend.
A set of markdown prompts, decision trees, and checklists that walk AI coding assistants through structured PRD, architecture, and security steps before they generate code.
A Django REST Framework backend with JWT authentication, providing CRUD APIs for patients, doctors, and doctor-patient records, backed by PostgreSQL. Built during my WhatBytes backend internship.
A FastAPI backend tracking per-user credits with a scheduled daily top-up job, also built during my WhatBytes internship.
A FastAPI blog backend with JWT auth, posts, comments, and likes, from the same internship.
Links customer contact records that share an email or phone number into a single unified identity.
Real-time chat using an embedded Redis server for message storage and pub/sub over Server-Sent Events.
A REST API that finds toll plazas between two Indian pincodes using Google Maps geocoding and an in-memory database of 2,387 toll plazas.
Simulates order creation, payment, shipping, and cancellation using the Observer pattern to log and alert on status changes.
A Django backend built around API security, rate limiting, and access-control concerns.
Trains and compares Logistic Regression, SVM, and Random Forest classifiers on a UCI phishing-website dataset, reaching about 97% accuracy with Random Forest.
For project inquiries and collaboration, get in touch.