AI Engineer & Full-Stack Developer

I build advanced query systems & LLM pipelines that ship to production—fast.

  • • 14+ years in IT • building AI search, automation, and workflow products
  • • AWS certified • Snowflake Core Pro • production-first engineering
  • • Full‑stack across FastAPI, Streamlit, Next.js, Electron, and MAUI
Works with
AWSAzure AIOpenAILlamaIndexPostgreSQL/pgvectorRedis
Saravanakumar Subramani portrait

About

I’m an AI engineer and full-stack developer focused on search, summarization, automation, and workflow tools that have to work in production. Over 14 years in IT, including 8+ years across AI/ML and NLP, I’ve built systems for healthcare, sales, and knowledge-heavy teams using Python, LlamaIndex, FastAPI, Streamlit, PostgreSQL/pgvector, and cloud AI services on AWS and Azure.

  • • 14 years overall • 8+ years AI/ML & NLP
  • • Certifications: AWS SAA, Developer, CCP • Snowflake Core Pro
  • • Stack: Python, LlamaIndex, LangChain, OpenAI, FastAPI, Streamlit, PostgreSQL/pgvector, Redis, Next.js, PyTorch, Scikit-learn, AWS Bedrock, Azure AI Services

Core Focus

  • • RAG on pgvector/Redis
  • • Hybrid retrieval & filters
  • • Prompt & tool orchestration
  • • Caching & evals
  • • AI Search & Indexing
  • • MLOps Pipelines

Delivery

  • • FastAPI microservices
  • • Streamlit prototypes → prod
  • • Docker & CI/CD
  • • Cloudflare, AWS/Azure
  • • Hugging Face Deployment
  • • GitHub Actions Automation

Services

Outcome‑focused engineering. No buzzwords, just shipped systems.

Advanced Query Systems (RAG)

  • • LlamaIndex on pgvector/Redis
  • • Hybrid retrieval, metadata filters

LLM Pipelines

  • • Prompts, tools, guardrails
  • • Evals, caching, observability

APIs & Microservices

  • • FastAPI, Streamlit
  • • Docker, CI/CD

Cloud & WordPress

  • • AWS/Azure, Cloudflare, Hetzner
  • • Perf & cost tuning
LlamaIndexFastAPIPostgreSQL/pgvectorRedisBedrock/OpenAIAzure AINext.js

Featured Projects

A mix of public GitHub builds and hands-on workflow projects across healthcare NLP, browser automation, speech, and practical AI tooling.

DealCloser assistant screenshot

Real-time sales copilot

DealCloser Agent Assistant

GitHub

Cross-platform assistant for sales and support teams that listens to live conversations, transcribes them in real time, and surfaces context-aware suggestions while the call is still happening.

  • • Azure Speech powers continuous speech recognition with live transcript updates.
  • • MAUI client and backend suggestion service support objection handling and next best action guidance.
.NET MAUIC#Azure SpeechREST APISales AI
Screen Recorder and Browser Automation screenshot

Desktop automation

Screen Recorder + Browser Automation

Product

Electron desktop app that records browser actions and desktop video, then turns sessions into step-by-step tutorials with screenshots and replayable automation.

  • • Captures clicks, typing, navigation steps, and tutorial screenshots during a session.
  • • Exports flows as Playwright scripts and stores tutorials in PostgreSQL.
ElectronPlaywrightPostgreSQLMediaRecorderDesktop App
Clinical summarization app screenshot

Healthcare NLP

Clinical Summarization App

GitHub

Clinical note processing workflow that transforms unstructured documents into structured summaries and supports chat-based review inside a Streamlit UI.

  • • Combines upload, summarization, and question-answer review in one interface.
  • • Service layer covers file processing, vector storage, and follow-up chat interactions.
PythonStreamlitNLPVector StoreHealthcare
Appeal letter automation screenshot

Healthcare automation

Appeal Letter Automation

Workflow

Workflow for generating consistent medical appeal letters from clinical documents and images with a human-review step before submission.

  • • Combines medical document understanding with structured output generation.
  • • Designed to reduce repetitive drafting work while preserving review control.
AWS Comprehend MedicalBedrockOpenAIS3Healthcare Ops
Workflow snapshot

Conversational interface

3D Talking Avatar POC

GitHub

Text-to-speech avatar system that synchronizes Azure Neural TTS visemes with a web-based 3D face for more natural AI interactions.

  • • FastAPI backend returns audio plus blendshape keyframes.
  • • React and Three.js frontend animates a fallback or custom GLB avatar.
PythonFastAPIReactThree.jsAzure TTS
AI Smart Web Reader screenshot

Browser AI tool

AI Smart Web Reader

GitHub

Browser extension that cleans up any webpage and lets users ask quick AI questions against the extracted content.

  • • Built as a Plasmo extension with an injected floating reader UI.
  • • Designed for fast, low-friction reading and question answering on the open web.
TypeScriptPlasmoBrowser ExtensionAI UX
View all GitHub repositoriesMore work includes LlamaDBIndexer, Azure AI Search REST Operations, PDF semantic search, and multi-agent experiments.

Timeline

A new blog post or app launch most weeks, plus 61 dated projects since 2018, and the skills each one took.

See the full timeline →

Latest

Eras

  1. 2018 – 2022

    Web3 & data engineering

    Ethereum and Flow smart contracts, Flask apps, and PySpark pipelines into Postgres and S3. Launched the Booksgrub online bookstore.

  2. 2023 – 2024

    Cloud AI & RAG foundations

    Medical text extraction on AWS (Rekognition, Comprehend Medical, Bedrock), then LlamaIndex + pgvector indexing and Azure AI Search.

  3. 2025

    Agents, healthcare NLP & open source

    MCP servers, multi-agent workflows, clinical summarization and semantic search, mostly published on GitHub, plus Tamil LlamaIndex tutorials.

  4. 2026

    Product studio & enterprise AI

    Shipping Digitalgrub apps to iOS, visionOS, Android, macOS, Windows and the web on a self-hosted fleet, with a blog post or launch most weeks, alongside client RAG and healthcare work.

Deep Case Study — Clinical Summarization

  • Problem: Manual review of unstructured clinical notes is slow.
  • Approach: Extraction + rules → standardized summaries; human‑in‑the‑loop.
  • Stack: Python, BiomedBERT, Streamlit, PostgreSQL.
  • Result: Review time ↓, edits ↓, consistency ↑ (validating).

Planned metrics (to instrument):

  • • Avg. review time per case
  • • Extraction coverage (% fields filled)
  • • Manual edits per summary
  • • Clinician agreement rate (spot‑checks)

Content

Let’s Chat

Book time or drop a message. I respond quickly.