✦ AI/ML ENGINEER // OPEN TO OPPORTUNITIES

Intelligence.
Engineered for Business Impact.

I design and build autonomous AI agents, computer vision systems, and high-precision machine learning pipelines — as a full-time AI/ML engineer or freelance collaborator — with a focus on production-grade, explainable results.

Get In Touch ↗ Explore Case Studies ›
Interactive Neural Core — Move cursor to interact with node tensors
-60%
Analysis Overhead
Multi-cloud forecasting engine cutting manual effort
95%+
Computer Vision Accuracy
MediaPipe pose tracking & rep estimation precision
10-Agent
Stateful Workflow DAG
LangGraph pipeline with automated risk scoring
8.2 CGPA
SPPU Data Science '25
Guinness World Record Contributor // Pune, India
Solutions & Offerings

Production AI built for your exact problem.

Whether you're hiring for a full-time seat or a focused freelance engagement, here's how I approach production AI — end-to-end, from model design to deployment.

FLAGSHIP CAPABILITY

Autonomous AI Agents & Workflows

Stateful, multi-agent systems built with LangGraph, Llama 3, and LangChain. Empower your product to reason across tools, query vector databases, and execute multi-step business workflows with robust human-in-the-loop safeguards.

  • Multi-Agent Consensus & Failover DAGs
  • Production RAG with Hybrid Semantic Search
  • Explainable Audit Reports & Tool Execution
  • Llama 3 / GPT-4 / Gemini Model Optimization
Available: Full-Time • Contract
EDGE & REAL-TIME

Computer Vision & Edge Systems

Real-time video inference and keypoint detection using OpenCV and MediaPipe. Deploy high-speed motion tracking, exercise rep counting, anomaly classification, and object detection running smoothly in the browser or on the edge.

  • Sub-20ms Real-Time Pose Estimation
  • Biomechanical Angle & Rep Counting Engines
  • Custom Object Detection & Feature Tracking
  • Lightweight Edge Model Deployment
Available: Full-Time • Contract
FINANCIAL & OPERATIONS

Predictive ML & Forecasting Engines

Automated time-series and risk assessment models combining Prophet, XGBoost, and Scikit-Learn. Turn complex historical data into reliable cost forecasts, risk scores, and executive dashboards that give leadership clarity.

  • Hybrid Prophet + XGBoost Ensemble Forecasting
  • Automated Risk Scoring & Proposal Screening
  • FastAPI Microservices + React Dashboards
  • Natural Language Insights via Local LLMs
Available: Full-Time • Contract
Proven Results

Featured Case Studies.

Explore shipped production systems that combine algorithmic rigor with high-impact business outcomes.

01 CLOUD FINANCIAL INTELLIGENCE

Cloud Cost Forecasting Engine

Designed an intelligent FinOps forecasting system that combines Prophet time-series modeling with XGBoost gradient boosting to predict multi-cloud infrastructure expenditure. Integrated Llama 3 to deliver conversational, natural language variance analysis.

-60% Manual Analysis Effort
Dual-Model Prophet + XGBoost Ensemble
React Flask Prophet XGBoost Llama 3 Agentic AI
View Repository ↗
FinOps Cost Prediction Dashboard
Prophet Trend XGBoost Residuals Actual Spend
Llama 3 Insight: Projected 18.4% AWS anomaly detected in EU-Central node.
02 AUTONOMOUS MULTI-AGENT PIPELINE

MIDC Proposal Approval Risk Predictor

Engineered an autonomous AI evaluation desk for industrial project proposals using a 10-agent LangGraph architecture. Automates multi-factor risk scoring, validates compliance bylaws, and generates transparent, explainable audit dossiers prior to human board review.

10 Agents Collaborative LangGraph Pipeline
Instant Compliance Audit Generation
Python LangGraph AI Agents Risk Assessment Scikit-Learn
View Repository ↗
10-Agent LangGraph Execution Graph
Ingest
→
Compliance
→
Financials
→
Risk Eval
→
Audit Dossier
✓ Ingestion verified: 48 PDF document pages parsed ✓ Zoning bylaws: 100% compliant with MIDC industrial codes ⚠ Debt-to-Equity ratio: 2.4 (Flagged for committee review) ★ Composite Risk Index: 0.28 (Low Risk Profile)
03 COMPUTER VISION & EDGE

Real-Time Workout Tracker

Developed a high-performance computer vision system for real-time exercise form tracking and automated repetition counting using OpenCV and MediaPipe. Calculates 3D joint angles on live camera feeds with over 95% detection accuracy.

95%+ Repetition Counting Precision
30 FPS Real-Time Edge Inference
Python OpenCV MediaPipe Pose Estimation Biometrics
View Repository ↗
MediaPipe 33-Landmark Pose Stream
ELBOW ANGLE:88.4°
REP COUNT:14 REPS
FORM ACCURACY:98.2%
Interactive Verification

Test the intelligence live.

Experience how my agentic workflows and predictive models behave in real time before booking a project.

Autonomous Agent Reasoning Inspector

A simulated LangGraph multi-agent workflow — watch it reason through a realistic task step-by-step.

TASK GOAL: "Audit cloud spend spike in EU-West cluster & generate remediation PR."
READY Agent orchestrator initialized. Click "Simulate Agent Task" above to run workflow.

Cloud Cost Impact Model (Demo)

An illustrative walkthrough of the calculation behind my Cloud Cost Forecasting Engine — adjust the slider to see it in action.

$15,000 / mo
MODELED ANNUAL SAVINGS $108,000 Based on the 60% manual-analysis reduction measured on my forecasting engine project
HOURS SAVED PER MONTH 48 Hours Freed engineering capacity for core feature delivery

Pose Estimation Playground

Drag the joints of my Workout Tracker's skeleton model — angles and rep counts update live, mirroring the real MediaPipe pipeline.

Best experienced with a mouse — try dragging, or on desktop for the smoothest feel.

0°
Joint Angle
180°
Rep Count
0

Drag the pink (elbow/knee) or outer (wrist/ankle) joint through a full range of motion to trigger a rep — just like the real rep-counting logic. Keyboard: Tab to a joint, then use arrow keys.

Working Style

See how I'd scope a project like this.

Select the technical requirements below to instantly generate a recommended architecture and typical delivery window — then pre-fill a message to me.

01 Select AI Domain & Core Capability
02 Select Engagement Scope
03 Select Deliverables
RECOMMENDED ARCHITECTURE LangGraph Multi-Agent + Llama 3 + FastAPI
ESTIMATED TIMEFRAME 2 – 3 Weeks
Discuss This With Me ↓
Interactive Arcade

Neural Agent Speed Run.

Pilot an autonomous AI agent through network nodes. Collect verified clean data tokens while avoiding hallucination anomalies to set the high score!

TOKENS PROCESSED 0
PIPELINE INTEGRITY
HIGH SCORE 0
Controls: Use Arrow Keys or W/A/S/D (or touch/drag on mobile) to steer your Agent Core. Green = Clean Tokens (+10), Red = Hallucination Anomaly (-25 HP).
Initiate Contact

Let’s build something extraordinary.

Have a project in mind, an agentic workflow to automate, or need an expert machine learning engineer? Let’s talk timeline, architecture, and deliverables.

DIRECT INBOX

deshmukhkunal556@gmail.com

Direct line to Kunal. Guaranteed response within 24 hours. +91 84462 01187

LOCATION & WORK AVAILABILITY

Pune, Maharashtra, India

Open to full-time AI/ML roles and select freelance/contract engagements, remote-first.

ENGINEER GUARANTEE

Explainable • Production-Ready • On-Time

Every deliverable includes automated unit tests, clean documentation, and post-deployment handover.