Chandan Kumar

AI/ML Engineer, ship models, not just notebooks.

Bengaluru, IN

Chandan Kumar

About

I build production AI systems across large language models, retrieval pipelines, and computer-vision tooling that move from notebooks to real users. Currently focused on LLM fine-tuning, RAG, and multimodal perception.

Final-year CSE student at VIT Vellore, with internships at ISRO, IIT Ropar, IIT Bombay, and AI startups. I obsess over the path from notebook to deployed system: latency, model size, retrieval quality, and the boring infra glue that makes models actually useful. I also send fixes upstream to the open-source ML stack I build on.

Work Experience

  1. OSVI.ai

    Founding AI Engineer

    Aug 2026 to Present
    • Building the core of OSVI's real-time voice AI platform: the agent engine that runs every call, from speech recognition and LLM reasoning to speech synthesis and telephony.
    • Built its voice-to-voice calling and speech stack, cutting response latency and keeping calls running through model outages.
  2. NeuroFin.ai

    AI Engineer

    Sep 2025 to Jul 2026Onsite · Bengaluru
    • Built an intelligent KYC verification system using computer vision and deep learning to automate document validation and identity matching.
    • Designed a robust pipeline that classifies document types, extracts key fields, and performs face matching for secure user onboarding.
  3. Dec 2025 to Apr 2026Remote · Mumbai
    • Designed and shipped ML systems at Turocrates, focused on production data pipelines and inference services.
    • Owned the full model lifecycle: training, evaluation, deployment, and observability across the platform.
  4. IIT Ropar · Annam.ai

    AI Research Intern

    Jun 2025 to Jul 2025Onsite · Punjab
    • Built a CNN plant-disease classifier reaching 93% accuracy via ResNet-50 transfer learning on 87K+ images.
    • Developed an ML recommendation engine and integrated NLP-based sentiment analysis for agricultural news.
  5. ISRO · LPSC

    Machine Learning Intern

    Jun 2024 to Jul 2024Onsite · Kerala
    • Created a custom OCR pipeline combining YOLOv5 and PaddleOCR to digitize 500+ engineering drawings.
    • Applied image preprocessing techniques that improved model robustness by 15%.

Open Source Contributions

15 merged PRs across 11 projects, ranked by the organization behind each project

Projects

Things I've built end to end

Open-source · LLM Security

A Python library and MCP server that detects and masks PII before it reaches LLMs. Hybrid detection engine combines regex with Microsoft Presidio and spaCy NER, using reversible tokenization and a per-request in-memory vault. Ships as a library, FastAPI middleware, CLI, and MCP server for Claude Code.

Approached by Apertu Capital (Don Sheu & Enrico) for funding.

PythonFastAPIPresidiospaCyMCP

Local LLMs · 1-bit Quantization

One-click 1-bit (IQ1_S) compressor for any local LLM on Apple Silicon. Auto-discovers Ollama and on-disk GGUF models, runs importance-matrix calibration, and quantizes via llama.cpp with Metal — shrinking Llama-3.1-8B from 16.1 GB to 2.19 GB (7.4×) at ~51 tok/s on 3.1 GB of RAM. Side-by-side playground streams live tokens/sec, RAM, and bandwidth.

PythonFastAPIllama.cppReactMetal

Computer Vision · Identity

Face verification pipeline that compares faces across images, PDFs, and Excel documents. RetinaFace detection with quality filtering and automatic rotation handling, ArcFace embeddings via DeepFace, and cosine-similarity matching with configurable thresholds — producing detailed reports with per-document confidence scores.

PythonDeepFaceArcFaceRetinaFaceOpenCV

Vocat

2025

Voice AI · Real-time

Real-time voice interview agent. Google-Meet-style UI with WebRTC audio, Whisper STT, GPT-4o reasoning, and ElevenLabs TTS. Streaming responses with VAD-based turn-taking for natural conversational latency.

ReactPythonWhisperElevenLabs

Speech · LoRA Fine-tuning

Fine-tunes OpenAI Whisper-small for Hindi speech recognition using LoRA (PEFT) on Mozilla Common Voice 17, training under 2% of parameters to measure WER/CER gains over the baseline. Type-hinted library with YAML-driven config, runs as fast CPU/MPS smoke tests locally and full fp16 training on a Colab T4.

PythonPyTorchWhisperPEFTHuggingFace

Publications

Research papers

Automated Ranking of Video Frames Based on Clarity

IEEE PICC 2025
Oct 2025 · PublishedIEEE Xplore

Achievements

Recognitions and highlights

Apertu Capital · Funding interest

2026

ShieldPrompt approached by Don Sheu & Enrico

Outreach Head · IEEE SPS, VIT

2024

Organized 5+ technical workshops

7th Rank · CodeChef-VIT Hackathon

Feb 2023

Top 2% among 400+ teams

Skills

CNNRNNLSTMGANTransformersTransfer LearningGPTMistral-7BBERTLangChainLlamaIndexRAGLoRAPEFTYOLOResNetPaddleOCRTesseractFace RecognitionSegmentationPythonGoTypeScriptJavaC/C++SQLAWSDockergRPCFastAPIPostgreSQLRedisMongoDB

Showing 33 total skills

Education

VIT Vellore

Computer Science and Engineering

Final year