Hi, my name is

Ayushmaan Das.

I build AI systems that ship.

AI/ML & Software Engineer specializing in LLMs, RAG, Agentic AI, Edge AI, and MLOps. Currently building production AI systems from inference to deployment at MulticoreWare.

Hero Image

01. About Me

I build AI systems that ship - from low-level model optimization on edge hardware to full-stack AI platforms serving real customers in production. My engineering focus spans building end-to-end MLOps pipelines, designing robust RAG architectures (embedding, re-ranking, retrieval), executing fine-tuning orchestration, and deploying multi-tenant adapter inference engines. I care just as much about the infrastructure layer as the models themselves. Using tools like AWS, Kubernetes, and Docker, I architect scalable backends featuring Stripe-based billing, LLM-token-based autoscaling, secure RBAC authentication, and comprehensive Prometheus observability to ensure production-grade reliability.


On the systems programming side, I write high-performance C++ for custom inference servers, leveraging asynchronous multi-threading and optimized KV Caching (achieving up to 33.71 tok/s). My expertise in Edge AI allows me to push hardware boundaries using state-of-the-art quantization and calibration techniques like LiteRT and AIMET. I recently applied these principles to architect an end-to-end AI-enabled IDE for embedded systems - featuring comprehensive model-optimization pipelines that take neural networks from raw import to hardware-ready deployment, alongside custom tooling like model visualizers, latency estimators, and debug dashboards.


Outside of my core engineering roles, I actively research and implement Agentic AI workflows, focusing on frameworks like LangChain, ReAct Agents, and the Model Context Protocol (MCP). My recent personal projects include a dual-pipeline medical diagnostic assistant combining fixed RAG with dynamic reasoning agents, and an emotion-aware Spotify recommendation engine driven by custom neural networks and LSTM-based generation. I graduated as the Founder Chancellor's Gold Medalist for my B.Tech in AI & ML, and I am always excited to tackle complex engineering challenges or talk shop on modern AI architecture.


Additionally, I am a 3x published book chapter author, with peer-reviewed contributions in the fields of Artificial Intelligence and Machine Learning.

02. Technical Arsenal

Languages

Python
C++
C
Java
SQL
NoSQL

AI/ML Tools

LangChain
Agentic AI
ReAct Agents
MCP
KV Caching
RAG
Fine-Tuning
TensorFlow
PyTorch
Keras
LiteRT
AIMET
MLFlow
Hugging Face
FAISS
SpaCy
Scikit-Learn
OpenCV
Sentence-Transformers
Quantization
Neural Networks
Deep Learning
Reinforcement Learning
MLOps
Prompt Engineering

Frameworks

Flask
FastAPI
Streamlit
Gradio
Node.js

Cloud & Infra

AWS (S3, EKS)
Kubernetes
Docker
Azure DevOps
Prometheus
CI/CD
Koyeb
OCI

Other Tools

Git / GitHub
Linux
MongoDB
Power BI
PlatformIO
APIs
IDE Development

03. Where I've Worked

Jul 2025 - Present

Software Engineer

MulticoreWare Inc.
  • Built an AI IDE for embedded/edge hardware, integrating a quantization and optimizer pipeline with model graph visualization, memory layout visualization, plotting tools, and layer-wise optimization.
  • Bridged the gap between IDE and hardware SDKs/toolchains, owning the effort end-to-end - architecting, proposing new technologies, and implementing at the hardware level.
  • Completely owned the development of the production-grade product, extending multi-OS support, testing, and monitoring releases to end-users.
  • Engineered on-device model optimization with LiteRT and AIMET quantization, shrinking model footprint and boosting inference throughput on edge hardware.
  • Built the MVP of an AI Playground platform end-to-end - RAG infrastructure (embedding + re-ranking), fine-tuning job orchestration, and adapter-based inference modules serving multiple tenants.
  • Architected core backend systems - RBAC, JWT-based auth, and file management - with security and input-validation layers hardened across every service endpoint. Handled monetization using Stripe.
  • Deployed and scaled the platform on AWS (S3 + EKS) with Kubernetes, setting up Prometheus monitoring and observability across microservices.
Jan 2025 - Jun 2025

Junior Engineer

MulticoreWare Inc.
  • Onboarded and optimized LLMs (CodeLlama-7B/13B, Codestral-22B, Qwen2.5-Coder-32B) for Qualcomm edge hardware in C++, enabling hardware-accelerated, Mxfp6-quantized inference with KV caching and layer-wise operator compatibility fixes, boosting throughput up to 33.71 tok/s.
  • Built a high-performance C++ web server exposing LLM inference endpoints - async processing, multi-threaded architecture, and implicit load balancing - supporting concurrent inference requests across multiple models.
  • Extended deployment coverage across heterogeneous environments, contributing to OCI cloud and edge-device-specific framework integrations spanning cloud and on-device targets. Worked with MLflow for tracking experiments.
Jun 2024 - Dec 2024

Software Development Intern (ML Team)

Upraised®
  • Designed an intelligent proctoring system using pyannote speaker diarization to detect and flag anomalous audio patterns during assessments.
  • Built a RAG-powered backend for automated assignment generation - recruiters could generate role-specific tech/non-tech assignments from natural language prompts.
  • Developed and maintained core AI platform features, integrating AI-driven automation to reduce manual recruiter workflows.
Feb 2023 - Oct 2023

ML and Python Developer Intern

RecruitNXT
  • Designed and implemented a complete resume parsing solution from scratch, enabling automated extraction of key candidate information.
  • Created and annotated custom datasets, built and fine-tuned Named Entity Recognition (NER) models using SpaCy for precise data extraction, deploying them using Docker.

04. Some Things I've Built

Medical Agentic RAG

Dual-pipeline diagnostic assistant for medical-field Q/A, utilizing LangChain ReAct orchestration with Llama 3.3 70B and FAISS-based RAG with a fine-tuned FLAN-T5 model.

RAG LLaMA FLAN-T5 LangChain ReAct FAISS TF Serving Docker

Spotify Music Recommendation Engine

Emotion-based and custom ML algorithm-based song recommendation web application integrated with Spotify API. Built custom CNN model from scratch using TensorFlow and Keras.

TensorFlow Keras Cosine Similarity CNN LSTM Streamlit Spotify API

GlossaCompiler

Conversational AI for code assistance integrating a custom Transformer model in PyTorch. Generates executable Python code from plain English prompts via a Discord chatbot and React/Flask web app, with live compilation via Judge0.

PyTorch Transformers React Flask Discord API Judge0

ANPR Dashboard & Chatbot

Flask application utilizing TensorFlow Object Detection and EasyOCR to extract vehicular number plates. Stores live data in Google Sheets, visualized via a PowerBI dashboard, and features an OpenAI-powered chatbot.

OpenCV OCR TensorFlow PowerBI OpenAI API Flask

Highway Traffic Analysis

Machine Learning web app utilizing the Twitter API to fetch and analyze real-time highway traffic data. Features predictive modeling using Random Forest, Logistic Regression, and Naive Bayes Classifiers.

Random Forest Naive Bayes Logistic Regression scikit-learn Twitter API Streamlit

Resume & Namecard NER

FastAPI-Gradio application automating the hiring process by extracting entities from resumes. Features a custom NER model built with SpaCy and Doccano, data stored in MongoDB Atlas, and a querying chatbot.

NLP SpaCy OpenCV Doccano FastAPI MongoDB

05. Recognitions & Research

Achievements

  • Spot Award

    MulticoreWare · 2026
    Recognized for exceptional ownership, technical proficiency, and a proactive problem-solving approach leading to smooth, timely feature deliveries.

  • Gold Medal for Best Outgoing Student

    SRIHER, Chennai · 2025
    Awarded for outstanding academic performance and achieving the highest CGPA in the graduating cohort.

  • Research Day Winner (Best Paper Presentation)

    SRIHER, Chennai · 2024
    Won top honors for presenting an emotion-based music recommendation engine integrated with the Spotify API.

Publications

  • Composite AI-Driven Music Recommendation

    Wiley · May 2026
    Composite Artificial Intelligence: Fundamentals, Challenges, and Applications

  • Streaming Highway Traffic Alerts using Twitter API

    Taylor & Francis Group · 2024
    Advancement of Data Processing Methods for Artificial and Computing Intelligence

  • Disseminating Dynamic Traffic Information for Sustainable Mobility

    Taylor & Francis Group · 2024
    Artificial Intelligence and Machine Learning for Smart Community

06. What's Next?

Get In Touch

I'm always open to discussing modern AI systems, model optimization, or production AI infrastructure. Whether you have a question or just want to say hi, my inbox is always open!

Download Resume Email Phone WhatsApp