Headline

Builder at the Intersection of AI/ML, Physics, Data Science & Real-World Systems | BITS Pilani × IIT Madras | IMD(Gov. of India) - VIMANA - Blueocean | Applied ML, Data Engineering, Full-Stack

About

I am a dual degree graduate from 𝐁𝐈𝐓𝐒 𝐏𝐢𝐥𝐚𝐧𝐢 and 𝐈𝐈𝐓 𝐌𝐚𝐝𝐫𝐚𝐬, with 𝟑+ 𝐲𝐞𝐚𝐫𝐬 of hands on experience building agentic systems, evaluation pipelines, and production AI/ML systems across freelance work, startups, and a Government of India research lab, owning projects end to end. My core strength is taking a system from a rough idea to something that actually holds up in production, a skill built through owning systems end to end rather than just contributing. Highlights, 1. 𝐈𝐌𝐃, 𝐆𝐨𝐯𝐞𝐫𝐧𝐦𝐞𝐧𝐭 𝐨𝐟 𝐈𝐧𝐝𝐢𝐚 𝐫𝐞𝐬𝐞𝐚𝐫𝐜𝐡 𝐥𝐚𝐛: Built 0-72hr weather prediction system stacking LSTM, Transformer, LightGBM and XGBoost. 2. 𝐕𝐢𝐦𝐚𝐧𝐚, 𝐝𝐫𝐨𝐧𝐞 𝐬𝐲𝐬𝐭𝐞𝐦𝐬 𝐬𝐭𝐚𝐫𝐭𝐮𝐩: Owned PyTorch inference module for weather adaptive flight stabilization, improving stability by 80% in strong crosswinds, and reverse engineered QGroundControl codebase, porting to JavaScript stack for browser based drone control. 3. 𝐁𝐥𝐮𝐞𝐨𝐜𝐞𝐚𝐧, 𝐛𝐚𝐜𝐤𝐞𝐧𝐝 𝐚𝐧𝐝 𝐬𝐞𝐜𝐮𝐫𝐢𝐭𝐲: Architected full stack background verification ERP with multi layered security including custom encryption and JWT auth. 4. 𝐎𝐮𝐭𝐥𝐢𝐞𝐫 𝐚𝐧𝐝 𝐎𝐩𝐞𝐧𝐂𝐥𝐚𝐰, 𝐟𝐫𝐞𝐞𝐥𝐚𝐧𝐜𝐞 𝐋𝐋𝐌 𝐰𝐨𝐫𝐤: Frontier LLM evaluation and alignment work, RLHF-style annotation, red-teaming across general-purpose models, 4 custom tool-use agents built and 15+ agents red-teamed against OWASP Top 10 for LLM Applications categories. 5. 𝐁𝐥𝐮𝐞𝐨𝐜𝐞𝐚𝐧, 𝐌𝐋 𝐄𝐧𝐠𝐢𝐧𝐞𝐞𝐫 𝐚𝐧𝐝 𝐆𝐞𝐧𝐀𝐈 𝐃𝐞𝐯𝐞𝐥𝐨𝐩𝐞𝐫: Architecting Generative AI HR Assistant, adapting the hybrid retrieval and multi-tier LLM fallback design from OmniRAG project into production, alongside HR Analytics platform using XGBoost and deep learning to flag flight-risk and burnout signals. My toolkit includes Python, PyTorch, LightGBM, XGBoost, HuggingFace Transformers, ONNX Runtime, Airflow, MLflow, FastAPI, Node.js, Docker, Redis, LangChain, RAG pipelines, YOLOv11, ViT, EfficientNet, C++17, Linux, and Git. I am open to full time roles in India and remote roles globally. I am targeting Applied ML, MLOps, GenAI Engineering, and Computer Vision roles. Email : saarthak.batra@gmail.com Github : https://github.com/SaarthakBatra Portfolio : https://saarthakbatra.github.io/portfolio Resume : https://drive.google.com/drive/folders/1VcqCDlFI_z6VsVQjX8nSh_VTUH4nnkth?usp=drive_link

Location

Gurugram

AI stack

antigravity

cursor

github copilot

hugging face

langgraph

obsidian

Work experience

ML Engineer and GenAI Developer at Blueocean International India Pvt. Ltd., Feb 2026 - Present · 7 mos, Gurugram, • Architecting a Generative AI HR Assistant (FastAPI, LangChain/LangGraph, Google Gemini, Pinecone, Next.js/TypeScript, SQLite, Docker), adapting the hybrid retrieval and multi-tier LLM fallback design from OmniRAG (see Projects). Currently in beta testing with 2-3 employees to automate onboarding, policy compliance training, and HR support for small businesses without a dedicated HR team • Building a standalone HR Analytics platform with XGBoost and deep learning models, using class weighted training and ensemble stack- ing to handle severe class imbalance in rare event prediction. Trained on a manually labeled dataset of 1,500 employees across 5-8 features (overtime hours, monthly income, job satisfaction, absenteeism, tenure) to flag flight risk, burnout risk, and upskilling needs, reaching 60-70% validation accuracy so far with the dataset still growing • Driving both workstreams independently as a fresher hire, handling dataset labeling, feature engineering, and system architecture end to end, within a 5-person tech team reporting directly to the CEO

AI/ML Engineer at Outlier - OpenClaw Project, Oct 2025 - Jan 2026 · 3 mo, Remote, • Delivered RLHF-style human feedback annotation, red-teaming, and rubric-based evaluation across general-purpose LLMs on Outlier (including the OpenClaw project), sustaining 10-15 tasks/cycle at 70+ hrs/week with a greater than 90% acceptance rate • Created 4 custom tool-use/function-calling and multi-step reasoning agents and red-teamed 15+ agents within simulated multi-step task environments (synthetic SQL databases, mock APIs), achieving a 100% break rate by exploiting incorrect tool selection, malformed function arguments, and multi-step plan derailment; findings mapped to OWASP Top 10 for LLM Applications categories including prompt injection, hallucination under ambiguity, tool-misuse, and context-window exploitation • Authored 15+ evaluation rubrics from scratch and conducted 40+ individual model evaluations across structured 3-5 LLM head-to-head comparison studies (including Claude and OpenAI models), benchmarking outputs against authored rubrics across reasoning, safety, and task-completion dimensions

Backend (Full-Stack) and Security Engineer Intern at Blueocean International India Pvt. Ltd., Mar 2025 - Oct 2025 · 8 mos, Gurugram, • Architected and delivered a full-stack background verification ERP from scratch (Node.js, Express.js, MongoDB Atlas, Vanilla JS), engineering scalable document ingestion and pipeline orchestration workflows that scaled operations from 4–5 to 30–40 verifications per month • Engineered a multi-layered security architecture comprising custom payload encryption middleware, JWT session authentication, bcrypt password hashing, IP surveillance, and automated session cleanup, enforcing zero-trust request validation, strict access control, and real-time network traffic monitoring across all admin endpoints • Designed a cloud-native document ingestion and retrieval pipeline (Google Cloud Storage + Multer) for heterogeneous document types — identity proofs, certificates, and academic records, improving traceability and maintaining end-to-end integrity across GCS and MongoDB Atlas storage layers

ML Engineer and Systems Software Developer Intern at Vimana, Jul 2024 - Dec 2024 · 6 mos, Pilani, • Built PyTorch inference module and telemetry data preprocessing pipeline for weather-adaptive ML/DL drone flight stabilization system, fine-tuning on ∼50 hr flight telemetry for edge deployment on onboard compute, achieving 80% stability improvement over standard PID baseline in crosswinds above 10–15 m/s • Developed 3D product demo platform (Angular, Three.js, GLTF/GLB, GSAP, Node.js, MongoDB) to simulate hybrid VTOL mechanics and flight modes with semantic, interactive visualization, adopted as company’s primary investor-facing interface • Reverse-engineered the QGroundControl (QGC) C++/QML codebase and ported it to a Java backend and JavaScript web client with WebSocket-based command and telemetry flows, enabling browser-based control for custom drone hardware

ML Engineer and Full-Stack Developer Intern at Indian Meteorological Department (IMD), MoES, Govt. of India, May 2024 - Jul 2024 · 3 mos, Pune District, • Architected a 0-72hr probabilistic rainfall & hazard prediction system stacking LSTM/Transformer architectures (PyTorch) with LightGBM/XGBoost and fusing multi-source data pipelines (IMD API, OGIMET SYNOP/METAR, MOSDAC, ERA5, ECMWF/GFS GRIB2), applying Non-homogeneous Gaussian Regression (NGR) post-processing and isotonic regression calibration to address severe class imbalance in rare-event prediction, achieving 70–80% forecast accuracy at 72-hour lead time across 5 meteorological parameters with uncertainty aware, class-weighted training for rare events including heatwaves and cold waves • Engineered an 18-component Apache Airflow-orchestrated data pipeline handling live multi-source ingestion (IMD API, OGIMET SYNOP/METAR, MOSDAC, ERA5, ECMWF/GFS GRIB2, NetCDF4), quality control and gap-filling using Python, xarray, GeoPandas, rasterio, and R, followed by NWP refinement and ensemble model output generation, establishing the feature engineering and ingestion architecture for downstream probabilistic ensemble modeling • Identified a critical gap in IMD’s manual forecast verification workflow and engineered a full-stack Forecast Verification Portal (vanilla JS + PHP) that fed scientist-verified forecasts back into pipeline for continuous model fine-tuning, reducing scientists’ manual verification effort by 80% and executing 8 automated forecast cycles daily

Full-Stack Developer at Metadrix IT Solutions, Jan 2023 - Dec 2023 · 11 mo, Remote, • Founded and independently operated a full-stack freelance practice (Metadrix), delivering 10-15 paid projects for 9 clients across India, the US, and Australia spanning landing pages, e-commerce, product-showcase, and hiring marketplace platforms, using React, Angular or Vanilla JS/HTML/CSS, on 2-8 week turnarounds with full code/IP ownership handed over on completion (live samples in portfolio) • Engineered elaborate 3D hero and product-showcase websites for multiple international clients using Three.js and GSAP-driven scroll animations, delivering polished, production-grade interactive experiences that served as client’s primary marketing and sales asset • Owned the complete business lifecycle solo end-to-end including client acquisition, requirement scoping, pricing negotiation, and delivery across 9 client engagements spanning 3 continents, operating as sole technical and business point of contact throughout

Education

Bachelor of Engineering (B.E.), Mechanical Engineering, Birla Institute of Technology and Science, Pilani

Bachelor of Science (B.S.), Programming and Data Science, Indian Institute of Technology Madras

Public links

github: https://github.com/SaarthakBatra