PRASHANT
0%
Available for opportunities

PrashantJadhav

//Data Scientist & AI/ML Engineer

Building intelligent systems at the intersection of Machine Learning, LLMs, and Time Series Analysis. 4+ years turning complex data into decisions that matter.

View Projects →Get In TouchRequest Resume
Resume available on request — connect for a tailored AI/ML case study and CV.
// Core Stack
PythonML/DLLLMsRAGTime SeriesAWSNLPGen AI
// Experience Snapshot
8+
AI/ML Cases Delivered
30+
Enterprise Data Pipelines
100%
Quality & Deployment Focus

Enabling AI-driven decision-making with production-grade models, automation pipelines, and data-driven storytelling.

// 01 — About

The Data Mind
Behind the Models

// Profile Overview

Building AI-first systems that combine forecasting, NLP, and automation to deliver measurable business outcomes.

I'm a Data Scientist with 4+ years of hands-on experience crafting end-to-end AI/ML solutions across forecasting, natural language processing, and generative AI. Currently based in Ahmedabad, I work at Let's Enkindle where I build predictive models and intelligent automation systems for real-world business impact.

My expertise spans the full ML lifecycle — from raw data wrangling and feature engineering to deploying production-grade models via REST APIs on AWS. I thrive in cross-functional environments where data-driven decisions move the needle.

Recently, I've been deep-diving into the LLM ecosystem — fine-tuning models like SQLCoder, building RAG pipelines with vector databases, and developing AI agents powered by the Gemini API.

// Domains of Expertise
Machine Learning & Predictive Modeling
Time Series Forecasting (ARIMA, Prophet, LSTM)
Large Language Models & Fine-tuning
Retrieval-Augmented Generation (RAG)
Natural Language Processing (NLP)
Data Pipeline & Automation Engineering
Deep Learning (ANN, CNN, TensorFlow, PyTorch)
AWS Cloud Deployment & API Integration
Data Visualization & Business Intelligence
// 02 — Experience

Professional
Journey

4+ years building data-driven products across two forward-thinking organizations.

2024 – Present
Data Scientist
LET'S ENKINDLE · Ahmedabad, India
  • Performed Time Series Analysis on large, complex datasets to uncover patterns and trends informing business decisions.
  • Developed predictive models using Prophet, ARIMA, SARIMA, Exponential MA, and LSTM neural networks for product price forecasting.
  • Built AI-powered document processing pipelines integrating Gemini API for unstructured data extraction and normalization.
  • Applied advanced data visualization and summarization to communicate key findings to stakeholders.
  • Conducted research to stay current with advancements in AI and data science; periodically retrained models with new data.
2021 – 2024
Data Scientist
LETS NURTURE · Ahmedabad, India
  • Analyzed large datasets to identify trends and patterns supporting data-driven business strategies.
  • Developed classification and regression models using logistic regression, decision trees, and neural networks.
  • Applied data visualization techniques (Matplotlib, Seaborn, Power BI) to communicate insights across teams.
  • Collaborated with cross-functional teams to translate analytical findings into actionable strategies.
  • Conducted research to stay current with evolving data science practices and ML methodologies.
// 03 — Projects

Featured
Work

A selection of projects spanning Gen AI, NLP, Time Series, and ML classification.

📋
Gen AI · Automation
S&S Statement AI Verification Tool
End-to-end AI system automating stock statement verification for a pharma company with 100+ super stockists and ~13,000 monthly reports. Extracted and normalized data from PDF, scanned images, CSV, and Excel using Gemini API, reducing manual effort and compliance errors significantly.
Generative AILLMsGemini APIPythonPDF/Image ProcessingData NormalizationAutomation
View Case Study →
🗄️
LLM · RAG
NLP-to-SQL Query Generation (Fine-Tuned LLM + RAG)
Built a system converting natural-language prompts into accurate SQL queries using a fine-tuned Defog SQLCoder model. Implemented schema-aware retrieval, vector search, and prompt optimization to reduce hallucinations and enable direct database integration.
LLM Fine-tuningRAGHuggingFaceSQLCoderVector DBQdrantChromaSQL
// Details coming soon
💬
Chatbot · NLP
AI-Powered Conversational Chatbot
Developed an intelligent chatbot with contextual understanding and multi-turn conversation support. Built a full RAG pipeline for dynamic knowledge retrieval including intent handling, vector search, and prompt engineering for accurate and up-to-date responses.
LLMsRAGHuggingFaceVector DBPrompt EngineeringAPI Integration
// Details coming soon
📈
Time Series · Forecasting
Allyl Chloride Price Forecast
Forecasted future prices of Allyl Chloride to support pharmaceutical business growth using advanced time series models including ARIMA, SARIMA, Prophet, and LSTM, with comprehensive data preprocessing and visualization pipelines.
PythonPandasTime SeriesARIMAProphetLSTMData Visualization
// Details coming soon
⚡
Time Series · Energy
Coal & Sulphur Price Forecast
Built forecasting models for both Coal and Sulphur commodity prices, helping manufacturing industries make informed procurement decisions. Applied data cleaning, trend analysis, and multiple forecasting algorithms for reliable price predictions.
PythonPandasSARIMAExp. SmoothingData PreprocessingVisualization
// Details coming soon
🔧
ML · Classification
Predict Type of Failure in Machine
Multi-class classification system predicting machine failure types from multi-feature sensor datasets. Applied various ML classifiers with feature engineering and model evaluation to support predictive maintenance workflows.
PythonScikit-learnClassificationFeature EngineeringData Visualization
// Details coming soon
// 04 — Skills

Technical
Arsenal

🤖Machine Learning
Scikit-learn / Classical ML92%
Deep Learning (TF / PyTorch)85%
Time Series Analysis90%
NLP & Text Mining82%
🧬LLM & Gen AI
LLM Fine-tuning (HuggingFace)85%
RAG Pipelines88%
Vector Databases (Qdrant/Chroma)82%
Prompt Engineering87%
💻Programming & Data
Python (NumPy, Pandas, SciPy)93%
SQL (MySQL, MongoDB)85%
Data Visualization (Matplotlib, PBI)88%
Flask / Streamlit / REST APIs80%
☁️Cloud & Infrastructure
AWS (Server & API Deployment)78%
Snowflake72%
Statistics & Hypothesis Testing84%
Model Deployment & REST APIs80%
// Tech Ecosystem
ProphetARIMALSTMTensorFlowPyTorchHuggingFaceGemini APIQdrantChromaStreamlitFlaskSnowflake
// 05 — Education

Academic
Foundation

Bachelor of Engineering
Pune University
2018 – 2021
✦ 7.21 CGPA — First Class
Diploma in Engineering
MSBTE
2015 – 2018
✦ 77.29% — First Class with Distinction
SSC (10th Standard)
MSBSHSE
2014 – 2015
✦ 79.40% — First Class with Distinction
// 06 — Contact

Let's Build
Something Together

✉️
Email
pgjadhav0001@gmail.com
📞
Phone
+91 7666308973
📍
Location
Ahmedabad / Mumbai, India
🗓️
Availability
Open to full-time & contract roles
Ready to make data work for you?

Whether you're looking for an ML engineer, a GenAI specialist, or a data scientist to own your entire analytics stack — let's talk. I'm currently open to new opportunities and would love to connect.

📋 Gen AI · Automation · Pharma
S&S Statement AI Verification Tool
// Pharmaceutical Stock Reconciliation Automation · End-to-End AI Pipeline
◈ Project Flow// End-to-end pipeline
01
🏭
Problem Identification
02
🗂️
Data Requirements
03
🔗
Dealer Pipeline
04
🤖
AI Extraction
05
⚖️
Comparison Engine
06
✅
Validation & Labeling
▲ Problem Statement// Business pain points
•Client is a pharmaceutical manufacturing company selling products Pan India through a network of 30 super stockists, 200+ stockists, and thousands of chemists. Sales are tracked division-wise by state, stockist name, and division name.
•Each month, ground staff submit stock reports in PDF format — totalling 13,000+ PDF files per cycle. Ground staff often print reports, photograph them, and convert to PDF, resulting in non-standardized, unstructured data.
•A manual team of 10–12 people was responsible for verifying each dealer report against the master distributor stock list — comparing closing quantities product by product, stockist by stockist.
•The entire verification cycle consumed 10–12 working days per month, causing delays in compliance reporting, business decisions, and month-end reconciliation workflows.
•Goal: Build an automated AI system that ingests all dealer reports in any format, extracts structured data, and verifies them against the master stock report — reducing the 10–12 day cycle to near real-time.
◉ Data Requirements// Inputs & sources
The system required two primary data sources: dealer stock reports submitted by ground staff across India (in PDF, PNG, JPEG, Excel, and CSV formats), and the master stock report maintained internally by the client. A key integration step involved collaborating with the client's third-party system developer to establish an automated data dump pipeline into an AWS S3 bucket in a standardized format and path structure, ensuring consistent ingestion regardless of source format.
◈ Approach & Methodology// Technical solution design
TASK 01
Dealer Data Ingestion Pipeline (S3 Connection)
Built a Dealer Connection Pipeline to collect and route dealer data from the client's third-party system into an AWS S3 bucket. Defined a standardized path and format convention so all incoming reports — regardless of origin — land in a predictable, processable location.
TASK 02
Multi-Format Document Handling
Dealer reports arrived in five distinct formats. Developed format-detection logic and individual conversion handlers for each:
PDF — Direct text extraction attempted using pdfplumber; fallback to OCR for scanned/image-based PDFs
PNG / JPEG — Image-to-text via pytesseract OCR pipeline
Excel / CSV — Pandas-based ingestion with column normalization and schema alignment
TASK 03
AI-Powered Data Extraction via Gemini API
Traditional OCR and PDF extraction methods (pdfplumber, pytesseract) failed to produce reliable structured output from scanned, photographed, or non-tabular PDFs. The key engineering decision was to integrate Google Gemini API as the primary extraction engine.
Designed prompt templates instructing Gemini to parse each document and return data in the exact schema matching the master stock report
Applied prompt engineering to handle edge cases: missing columns, merged cells, handwritten annotations, and multi-page reports
Iteratively refined prompts through testing on diverse document samples from different ground staff
Output: Structured JSON/tabular data aligned with master stock schema — ready for comparison
TASK 04
Data Normalization & Schema Alignment
After extraction, applied a normalization layer to standardize product names, units, and division labels across all dealer reports. This step ensured extracted data was comparable with the master stock data despite variations in terminology and formatting used by different ground staff across states.
◈ Comparison Engine// Reconciliation logic
TASK 05
Filtered Matching & Primary Key Assignment
Implemented a multi-key filtering strategy to narrow comparison scope before running validation:
Filtered records by Stockist Name, Division, and Date
Applied Product Name as the primary key for joining dealer data against master data
Performed Closing Quantity comparison between extracted dealer values and master stock values
TASK 06
Two-Layer Validation & Result Labeling
Designed a two-layer validation system to handle all mismatch scenarios intelligently and produce business-readable labels:
Layer 1 — Product Resolution: If product not found in master data, system cross-checks unit details and re-attempts product name matching
Layer 2 — Division Check: If product name and unit verify correctly but quantity still mismatches, it is labeled as "Division Mismatch" — isolating the error type for the business team
// Output Labels
✓ Match✗ Not Match⚠ Division Mismatch◌ Product Not Found
// Layer 1 — Product Resolution
When a product is not found in master data, the system checks unit details and re-validates the product name. This handles naming discrepancies between ground staff terminology and master records.
// Layer 2 — Division Check
If product name and unit both verify correctly but quantity still mismatches, the system flags it as "Division Mismatch" — providing the business team with a precise, actionable error category.
◈ Tech Stack Used// Tools & technologies
ProphetARIMASARIMALSTMXGBoostTensorFlowPyTorchKerasNLTKHuggingFaceGemini APIQdrantChromaPower BIStreamlitFlask
◈ Key Outcomes & Results// Business impact
⚡
~90%
Reduction in manual verification time
👥
10–12
Person manual team replaced by automation
📄
13,000+
Monthly PDF reports processed automatically
🗓️
10–12d → hrs
Verification cycle drastically reduced
🎯
Multi-layer
Validation with business-readable error labels
🏭
Pan India
Coverage across all 30 super stockists & divisions
// Pharmaceutical · Stock Verification · AI Automation · 2024–Present