Initializing Portfolio…
AI · Live & listening

Vivek PolAI Engineer

Six years turning data and language models into products that ship — RAG, agents, voice, and the quiet infrastructure underneath.

Explore the work→Get in touch
Email[email protected]WhatsApp · Mob+91 94221 21962chat →
RAG · retrieval
Agents · tools
Voice · realtime
Vivek Pol
Scroll
About

AI/ML full-stack developer with 6+ years of turning models into products that survive production.,

Skilled in Machine Learning, Deep Learning, NLP and Generative AI. Hands-on with Python, statistical analysis, financial and business data processing, and scalable model deployment. Strong background in financial domain workflows, RAG, PDF and document extraction, data pipelines, cloud deployment on AWS, and decision-support analytics dashboards.

Vivek Pol, AI/ML Full-Stack Developer
subject · live
Vivek Pol
AI/ML Full-Stack Developer · Mumbai, India
onlinerec · 001

Generative & Agentic AI

RAG pipelines, LLM fine-tuning, LangChain, LangGraph, CrewAI, MCP, vector stores (FAISS, Pinecone, Chroma, Weaviate, Milvus), evaluation with RAGAS.

Classical ML & DL

TensorFlow, Keras, PyTorch, XGBoost, LightGBM, CatBoost, MLOps with MLflow / DVC, feature engineering and hyperparameter tuning.

Backend & Data

FastAPI, Flask, Streamlit, Gradio, PostgreSQL, MongoDB, MySQL, Kafka, SQLAlchemy, Pyspark, Dask, and PDF pipelines end-to-end.

Cloud & DevOps

AWS EC2 / Lightsail / Bedrock, Docker, Kubernetes, Helm, Terraform, Jenkins, Ansible, GitHub Actions, Keycloak, Razorpay.

Journey

Where I am today — and how I got here.

Six years, five roles, one throughline: turning models into products that ship.

  1. Jan 2025 — Present

    Python Developer

    ValueAdd Research & Analysis Solutions LLP

    Conversational RAG + NL2SQL agentic system across MySQL / Excel / PDF / TXT, full CI/CD via Jenkins + Terraform + AWS + Docker, Keycloak auth and Razorpay payments, Anthropic-powered microservices behind FastAPI / Node.js on Kafka + MongoDB, serving 100+ subscribers. Also shipped CVPerfect — an LLM-driven resume-screening system that cut HR screening time and cost ~90%.

  2. Jul 2023 — Sept 2024

    Development Officer

    LIC India

    Acheived Class-2 Central Government Officer Post, Cleared the LIC ADO Class-2 Officer competitive exam with All-India Rank 120. Alongside the role, built an ML fraud detection system (feature engineering, tuning, cross-validation) and an insurance-premium prediction system deployed for personalized quoting (Scikit-Learn + LSTM).

  3. Nov 2019 — Nov 2021

    Data Scientist

    Capstone Securities Analysis, Pune

    Stock-price prediction with Python + Flask + BeautifulSoup, credit-risk scoring with XGBoost + SHAP on Power BI, an email-intent classifier + auto-reply generator with Transformers, and a customer-support multi-agent system (LangGraph / CrewAI, FAISS/Chroma, FastAPI, React).

  4. Jul 2019 — Nov 2019

    Trainee Engineer

    Oriental Rubber Industries, Pune

    Smart conveyor-belt condition monitoring with vibration, temperature and motor-current sensors. Predictive maintenance system for rubber mixers, rollers and presses using Python, Scikit-learn, IoT sensor streams and SQLite.

  5. Oct 2018 — Apr 2019

    Data Analyst

    Volkswagen Pvt. Ltd., Pune

    Built customer churn prediction (Scikit-Learn, XGBoost), Power BI dashboards for supply-chain optimization, an ANPR system (OpenCV, YOLOv8, EasyOCR), loan-default prediction with SHAP explainability, and an early multi-agent workflow for document analysis (CrewAI / LangChain, pdfplumber, ChromaDB).

Skills

The toolkit, without progress bars.

Each module is a tool I've shipped with, not a bar I could fill in. Hover for context.

Programming
Python
6+ yrs · primary
Primary language across every project on this page.
Programming
FastAPI
3 yrs · Talk2Data + microservices
Default choice for anything Python + HTTP.
Programming
Flutter / Dart
1 yr · mobile builds
One codebase for Android + iOS surfaces.
Programming
SQL
5 yrs · NL2SQL, ETL
Windows, CTEs, and query plans I can actually read.
Frontend
JavaScript
3 yrs · wrappers + microservices
Glue between the model and the browser.
Frontend
HTML
5 yrs · every UI I ship
Semantic markup — the base of every accessible page.
Frontend
CSS
5 yrs · design systems
Layout, motion, custom properties.
Frontend
React
2 yrs · internal tools + wrappers
Components, hooks, state — the workhorse.
Frontend
Next.js
1 yr · this portfolio + wrappers
App Router, server components, image pipeline.
Frontend
Node.js
2 yrs · Talk2Data microservices
The service layer behind the LLM.
Backend
Streamlit / Gradio
3 yrs · demos + internal tools
Ship an ML demo in an afternoon, not a week.
AI
Advance RAG
2 yrs · Talk2Data + docs
Hybrid retrieval, query rewriting, rerankers, citations.
AI
LangChain
2 yrs · CVPerfect + more
Used where it earned; rewound where it got in the way.
AI
LangGraph
1 yr · multi-agent
Typed stateful graphs for real production agents.
AI
CrewAI
1 yr · agent teams
Team-of-agents pattern for research + doc analysis.
AI
n8n
1 yr · automations
Low-code automations to prototype agent flows fast.
AI
MCP
<1 yr · tool bridging
Model Context Protocol for portable tool interfaces.
AI
Qdrant
1 yr · vector store
Production vector DB — filters, payload search, hybrid.
AI
FAISS / Pinecone / Chroma / Weaviate / Milvus
2 yrs · RAG pipelines
The full menu of vector stores; picked per workload.
AI
OpenAI Embeddings
2 yrs · RAG default
text-embedding-3-* for most retrieval use-cases.
AI
BGE Reranker
1 yr · RAG re-ranking
Second-stage rerank to lift precision at k.
AI
AI Avatar
<1 yr · realtime demos
On-screen avatar synchronised to a live speech pipeline.
AI
Hedra
<1 yr · avatar video
Photoreal avatar video generation.
AI
Bey AI
<1 yr · realtime avatars
Realtime avatar rendering for voice agents.
AI
Speech ↔ Text ↔ Speech
1 yr · voice agents
STT + LLM + TTS with interruption + barge-in.
AI
Cartesia
<1 yr · streaming TTS
Sub-second, natural-sounding streaming TTS.
AI
Deepgram
<1 yr · streaming ASR
Low-latency streaming ASR with endpointing.
LLM
Anthropic Claude
2 yrs · Talk2Data core
Primary reasoning model for production systems.
LLM
OpenAI GPT
3 yrs · across products
GPT-4o / GPT-4.1 for reasoning + tool use.
LLM
DeepSeek
1 yr · CVPerfect
Reasoning at a fraction of the cost.
LLM
xAI Grok
<1 yr · evaluation
Alternative reasoning for eval + comparison.
LLM
Perplexity
<1 yr · research
Search-augmented reasoning as a research tool.
LLM
Google Gemini
<1 yr · multimodal
Long-context + native multimodality.
LLM
Kimi K3
<1 yr · long-context
Long-context reasoning for document workflows.
LLM
GLM-5.2
<1 yr · multilingual
Open-weight bilingual reasoning model.
LLM
Gemma
<1 yr · local fine-tunes
Compact open-weight model for on-device use.
LLM
MiniMax M3
<1 yr · multimodal exp.
Multimodal generation experiments.
LLM
Llama
1 yr · self-hosted
Self-hosted open-weight baseline via Ollama.
LLM
Ollama
1 yr · offline demos
Local LLM inference for privacy-first prototypes.
ML
Linear Regression
5 yrs · baselines
Every project starts with a linear baseline.
ML
Logistic Regression
5 yrs · classification
Classifier baseline + interpretable coefficients.
ML
Decision Tree
4 yrs · explainable ML
When you need a model a stakeholder can read.
ML
Random Forest
4 yrs · churn, credit risk
Bagging that still beats deep nets on tabular data.
ML
Gradient Boosting
4 yrs · competitions + prod
The tabular workhorse.
ML
XGBoost
4 yrs · credit risk, churn
The reliable choice when speed and accuracy matter.
ML
LightGBM
3 yrs · wide feature sets
Faster training on wide tables than XGBoost.
ML
CatBoost
2 yrs · categorical-heavy data
Native handling of categorical features.
ML
Scikit-Learn
6 yrs · every ML project
The Swiss army knife of classical ML.
ML
TensorFlow / Keras
4 yrs · CV + tabular
Model architecture, training loops, custom heads.
ML
PyTorch
3 yrs · research + fine-tuning
Fine-tuning, distillation, custom loss functions.
ML
MLOps (MLflow / DVC)
2 yrs · reproducible training
Experiment tracking + data versioning done properly.
Observability
LangSmith
1 yr · LLM traces + evals
Projects, runs, traces, evals for LLM apps.
Observability
RAGAS
<1 yr · RAG eval harness
Automated retrieval + generation quality scoring.
Observability
Postman
4 yrs · API exploration
Every API gets a Postman collection before code.
Observability
pytest
4 yrs · service test suites
Fixtures, parametrization, coverage as a discipline.
Observability
Selenium
3 yrs · browser tests + scraping
End-to-end browser tests + scraping when APIs are absent.
Observability
Prometheus
1 yr · service metrics
Metric collection for AI microservices.
Observability
Grafana
1 yr · dashboards
Dashboards for latency, tokens, cost per request.
Observability
Power BI
3 yrs · decision dashboards
Decision-support dashboards for finance workflows.
Data
Pandas / NumPy / SciPy
6 yrs · every project
The bread-and-butter of every notebook I open.
Data
PySpark / Dask
2 yrs · large tables
When Pandas runs out of RAM.
Data
PostgreSQL / MySQL
5 yrs · services
JSONB, partial indexes, materialized views.
Data
MongoDB
3 yrs · Talk2Data
Document store for conversation state + logs.
Data
Kafka
2 yrs · event backbone
Event backbone for multi-service AI apps.
Data
pdfplumber / PyMuPDF / Camelot
3 yrs · RAG intake
Tables + text extraction that survives messy PDFs.
Data
pytesseract / EasyOCR
3 yrs · ANPR, docs
OCR the boring way and the accurate way.
CV
OpenCV
4 yrs · ANPR + preprocess
Everything from tesseract prep to real-time capture.
CV
Ultralytics YOLO
3 yrs · ANPR, detection
Number-plate detection in traffic streams.
CV
MediaPipe / TorchVision
2 yrs · pose + face
Lightweight vision graphs for interactive demos.
Cloud
AWS (EC2 / Lightsail / Bedrock)
4 yrs · deployments
Compute, storage and hosted foundation models.
DevOps
Docker
4 yrs · 30+ services
Multi-stage builds, slim images, compose stacks.
DevOps
Kubernetes / Helm
2 yrs · internal ops
For services that need to survive a node dying.
DevOps
Terraform
2 yrs · AWS infra
Every environment is code, or it is not real.
DevOps
Jenkins
3 yrs · CI/CD
Pipelines from PR to prod, with tests along the way.
DevOps
Ansible
2 yrs · server provisioning
Idempotent playbooks for anything I SSH into.
DevOps
GitHub Actions
3 yrs · 20 pipelines
Build, test, deploy, evaluate models on every PR.
BI
Matplotlib / Seaborn / Plotly
5 yrs · analysis
The right chart for the audience, always.
BI
Tableau
2 yrs · dashboards
For stakeholders who live inside Tableau.
Auth
Keycloak
2 yrs · Talk2Data
Realms, SSO, RBAC for internal AI apps.
Auth
Razorpay
1 yr · Talk2Data
Subscriptions, webhooks and refund flows.
YouTube@vivek_pol
Vivek Pol

Everything I’ve explained in public.

YouTubeSubscribe
Open source

Where I think out loud, in commits.

Every one of my 85 public repositories, each tile linking straight to GitHub.

@polvivek77
last 6 months
Contribution rhythm
lessmore
Top languages
Python68%
Jupyter Notebook18%
HTML / CSS6%
Shell4%
Other4%
Repositories
85+ public repos
Learning

Formal ground, continuously extended.

Mechanical engineering foundation. AI/ML built on top through structured programmes and years of shipped systems.

  • June 2013 — June 2017

    B.E. Mechanical Engineering — First Class

    P.E.S. Modern College of Engineering, Pune

    HSC 69.16% · SSC 92.36%

  • 2024

    Full-Stack Data Science

    CodeWithHarry

    Structured programme covering ML, DL and productionisation.

  • 2024

    Generative Artificial Intelligence

    Udemy

    GenAI foundations, prompting, RAG, LangChain, evaluation.

Signals

Milestones I actually show up for.

  1. Talk2Data goes live

    2025

    Conversational RAG + NL2SQL system serving 100+ paying subscribers.

  2. CVPerfect — 90% reduction in HR screening time

    2024

    AI resume screener adopted by internal HR workflows.

  3. Generative AI (Udemy)

    2024

    Certificate in Generative AI — foundations, RAG, evaluation.

  4. Full-Stack Data Science (CodeWithHarry)

    2024

    End-to-end DS/ML programme from data to deployment.

  5. All-India Rank 120 — LIC ADO

    2023

    Central Government Class-2 Officer competitive exam.

  6. Multi-agent doc analysis in production

    2021

    CrewAI / LangChain pipeline for structured document review.

Get in touch

Let’s build something amazing together.

Voice agents, RAG systems, autonomous workflows, ML infra. If it is AI and it needs to ship, I’m the person to call.

Email
GitHub
LinkedIn
YouTube
X
Resume
Connect→
Email[email protected]WhatsApp · Mob+91 94221 21962chat →
Try typing python · coffee · sudo hire vivek