ML Engineer · Open to opportunities

AkshatSanghvi

Building production-grade ML systems.
LLM, RAG, voice AI, and everything in between.

See Work Get in Touch
Scroll
Experience

Where I've
Worked

Jun 2025 – Present
Full-time

Machine Learning Engineer

at Timble AI — New Delhi  ·  FinTech
  • Design and deploy production-grade ML and Generative AI systems for fintech workflows, aligning model behavior with business, compliance, and data privacy requirements.
  • Own end-to-end LLM-powered conversational systems: prompt engineering, RAG pipelines, multi-turn dialogue management, and structured outputs.
  • Build real-time voice and text inference pipelines using Django ASGI, WebSockets, Silero VAD, and Celery for async processing.
  • Deploy and operate containerised services on AWS/GCP, optimising for latency, reliability, and cost.
  • Work directly with enterprise clients, requirements gathering, demos, iteration cycles, and go-live readiness.
LLMs RAG Django ASGI AWS / GCP Celery
Dec 2024 – Apr 2025
Internship

Machine Learning Intern

at Salesken — Bengaluru, Karnataka
  • Fine-tuned Wav2Vec2 for English–Hindi ASR, replacing Whisper to improve transcription quality for multilingual sales calls.
  • Built end-to-end speech-to-text pipelines, noise filtering, 16kHz normalisation, and custom tokenisation.
  • Evaluated models with JiWER, achieving low Word Error Rate (WER) on Hinglish datasets.
  • Contributed to modular ML systems supporting summarisation, analytics, and downstream integrations.
ASR Wav2Vec2 PyTorch NLP
Projects

Selected
Work

View ↗
01

Cobra.AI – Multi-Agent RAG

Multi-agent RAG platform using FastAPI, LangChain & Gemini 2.5 Flash for collaborative analysis and real-time file querying.

FastAPILangChainGeminiRAG
View ↗
02

YouTube Automation with n8n

End-to-end automated content pipeline leveraging n8n, DeepSeek, Google Sheets, and YouTube API.

n8nDeepSeekYouTube API
View ↗
03

Football Match Analysis – YOLOv5

CV pipeline using fine-tuned YOLOv5 for player/ball detection, speed and distance estimation from match footage.

YOLOv5OpenCVFlask
View ↗
04

HIMYM NLP Analysis

Web app analyzing themes and character networks in "How I Met Your Mother" using NLP.

NLPNetworkXStreamlit
View ↗
05

Movie Recommender System

Content-based filtering using cosine similarity to suggest films based on user preferences.

scikit-learnPandasTMDB API
View ↗
06

Rice Disease Classifier

CNN classifier for rice crop diseases across 6 classes, trained on a labeled agricultural dataset.

CNNTensorFlowKeras
View ↗
07

Sentiment Analysis

Binary sentiment classifier for movie reviews using NLTK and traditional ML approaches.

NLTKscikit-learnPython
View ↗
08

Neural Network From Scratch

Full NN built with only NumPy and maths that classifies MNIST handwritten digits.

NumPyMathMNIST
Skills

Tech
Stack

Programming

Python SQL C C++ Data Structures & Algorithms

Machine Learning

Classification Regression Clustering Feature Engineering Model Evaluation

Classical ML

XGBoost Random Forests Support Vector Machines scikit-learn

Deep Learning

CNNs Transformers PyTorch TensorFlow Fine-tuning Wav2Vec2

Generative AI

LLMs Prompt Engineering RAG Agentic Workflows LangChain LangGraph

NLP

Intent Detection Text Classification Dialogue Management ASR

Frameworks

FastAPI Flask Django scikit-learn

MLOps / DevOps

Docker CI/CD Celery ML Pipelines Model Deployment AWS GCP

Databases

PostgreSQL MySQL MongoDB FAISS Pinecone

Domain

FinTech Systems Client-Facing AI Voice AI Multimodal AI
Highlights

Achievements

Open-source contributor to Mesop, a UI framework developed by Google.

5-star rating and 1st position in Python on HackerRank.

Successfully completed competitive data science and machine learning challenges.

B.Tech in Computer Science from Parul University, Vadodara | 2021–2025.