「 WANTED: AI/ML ENGINEER 」
AI / ML Engineer
I build production-grade AI systems — agentic LLM pipelines, deep learning models, and real-time data engines that actually ship, not just notebooks.
WANTED
DEAD OR ALIVE
MEKA D. VARDHAN
₿3,000,000,000-
Wanted for shipping production-grade AI. Armed with autonomous agents; considered extremely persuasive with LLMs. Approach with job offers only. Report all sightings to the nearest recruiter or Marine outpost immediately.
MARINEPIRATE PROFILE
tap to flip back ⟲
Chapter 01 · Captain's Log
In which our captain introduces himself, and the log begins.
I'm an AI/ML engineer who designs and ships production-grade AI systems — deep learning models in PyTorch and TensorFlow, and LLM pipelines wired into live streaming data, structured outputs, and safety gates.
I built Quintyx, a multi-agent LLM trading framework with live WebSocket ingestion, multi-provider fallback routing, schema-validated outputs, and confidence-based risk gating. My applied background spans Transformer-based NLP, RAG, semantic search, and explainable ML — backed by full-stack skills and clean, testable code.
Like any good pirate, I'm chasing the ultimate treasure: AI systems that never break mid-stream. 🏴☠️
Chapter 02 · The Voyage
Three seas crossed, and the map keeps growing.
FlyRank Corp.
Completed FlyRank's AI engineering cohort: 15 assignments across data wrangling, embeddings and clustering, intent and opportunity modeling, and insight to action. Capstone — "Google Search Ranking & Discoverability" — reviewed and accepted by the lead track mentor.
recommendation letter ↗ · final evaluation ↗SwiftSafe Cybersecurity Technologies
Built ML components for anomaly detection, risk scoring, and security-monitoring pipelines running on real operational data. Developed Python backend services and data-processing workflows for cybersecurity automation within a distributed remote team.
internship certificate ↗Self-Directed
Designed and built Quintyx, a multi-agent LLM trading framework — async indicator engine, provider fallback routing, JSON-schema validated outputs, and a FastAPI/WebSocket service layer. Went deep on the agentic stack: MCP, local inference (Ollama), vector retrieval (FAISS, ChromaDB), and LLM orchestration and evaluation patterns.
github.com/mdsvr/Quintyx ↗VIT-AP University
CGPA 7.65 / 10. Senior design project: AI-powered cost analysis system for textile production with XGBoost + genetic algorithm optimization.
Chapter 03 · Legendary Bounties
The treasures claimed so far — each with a price on its head.
Autonomous AI stock-analysis agent for Indian (NSE/BSE) & US markets — ingests live data, calls tools, reasons over signals, and returns gated BUY/SELL/HOLD decisions.
— HOW IT WORKS —
data_layer/Fyers WebSocket streams NSE/BSE
& US ticks — push, not polling.async_engine.pyEMA · RSI · MACD ·
ATR · OBV computed in parallel — 5–20× faster than sequential.fusion_engine/Hurst-exponent regime detection
plus a 0–100% confidence score.guardian.pySignals
below the ATR / confidence threshold are blocked. Full stop.reasoning_engine/Gemini 2.0 Flash →
OpenRouter → offline deterministic fallback; validator.py
schema-checks every output.api_layer/FastAPI + WebSocket pushes
each verdict to the browser tick-by-tick.Verdict: BUYSELLHOLD — but only if the Guardian opens the gate.
Ranks 12,023 content items so one editor with ~50 review slots opens the right pages first — decision support, never automated action.
— HOW IT WORKS —
w03_data_contract.ipynbNine exclusion
categories — label components asserted out before anything is fit.w05_model.ipynb8-feature logistic regression,
picked from 16 configs under nested CV — the smallest one, because no config won
stably.w06_validation_audit.ipynbClient-grouped folds:
adversarial validation hits 0.961 AUC, so a random split would grade memorisation.
0.631 vs 0.577 AUC; p@50 0.90 vs 0.68.w07_action_playbook.ipynbSeven ranked
actions, a stop rank, a human in front of every item. Two claims that failed
clustered resampling ship as retracted.AI-driven cost analysis system using XGBoost with genetic algorithm optimization — 94.5% accuracy, R² = 0.94 — plus SHAP-based explainability. Senior design project, VIT-AP.
— HOW IT WORKS —
dataset.pyFabric, category, brand tier,
manufacturing & transport costs, margins, tax — the full ledger in.XGBClassifier.ipynbXGBoost tuned with a
genetic algorithm — 94.5% accuracy, R² = 0.94.output.pyPredicted selling price with a
cost-wise breakdown, SHAP-explained.LSTM and Transformer-based multilingual text classification supporting 20+ languages at 95%+ accuracy — architectures implemented from research papers as working models.
— HOW IT WORKS —
Language Detection.csvLabeled text across 20+
languages, cleaned and tokenized.codee.pyLSTM and Transformer classifiers
built from the papers, trained head-to-head.Full-stack platform (React.js, Node.js, MongoDB) with REST APIs enabling sub-second pricing responses, integrating XGBoost + SHAP for dynamic, explainable pricing.
— HOW IT WORKS —
UI/Separate React apps for consumers
and sellers — a two-sided marketplace.server/Node/Express REST APIs serving
sub-second pricing responses from MongoDB.ML/Python XGBoost sets each price
dynamically, with SHAP explaining every adjustment.Chapter 04 · Devil Fruit Arsenal
An arsenal gathered from every island along the route.
Chapter 05 · Poneglyphs
Stone tablets that prove the journey happened.
Final Chapter · Join the Crew
Every great crew started with a single invitation.
— SBS CORNER —
Q: "Vardhan-san!! Are you actually open to AI/ML roles and collaborations right now?!" — P.N. Recruiter-san, Grand Line
A: Yes!! And unlike most SBS questions, this one gets a serious answer — the buttons below reach me directly. I reply faster than a News Coo. 🗞️