Resume

Ankit Bari

Ankit Bari Resume

PDF document

Download

Ankit Bari

bariankitvinod@gmail.com | linkedin.com/in/aijadugar | github.com/aijadugar

Education

Bachelor of Technology in Artificial Intelligence and Data Science2023 – 2026
Vidyavardhini's College of Engineering and Technology, Mumbai UniversityCGPA: 9.10 / 10.0
Diploma in Computer Engineering2021 – 2023
Theem College of Engineering, MSBTEPercentage: 85.3%

Work Experience

Solo FounderDec 2025 – Present
SiteVPN & SkyBoy · Mumbai, India
  • •SiteVPN, an open-source privacy solution offering temporary emails, phone numbers, and VPN routing for secure AI tool usage.
  • •Building SkyBoy, an open-source skill infrastructure (SKILL.md) for AI agents, via web catalog, CLI installer, and MCP server.
AI Research InternJun 2025 – Nov 2025
The SMM Hub · Mumbai, India
  • •Developed AI Chatbot for lead tracking, achieving 95% accurate user data capture.
  • •Implemented fintech authentication backend, achieving low-latency (150–200 ms) request handling for 100+ daily users.
  • •Built a multi-agent blog generation system using CrewAI for a client, orchestrating specialized agents (research, drafting, editing, image generation) via OpenRouter LLM APIs.

Projects

FermesLive | Code
Node.js · Express · Next.js · Sarvam AI · LiveKit · Google Places API · Supabase (OAuth) · Redis · Docker
  • •Engineered an AI land-suitability agent, orchestrating Sarvam's LLM with tool-calling over cited geospatial data (elevation, flood zone, soil) and Google Places for nearby suppliers.
  • •Delivered a derived-verdict site report grounding LLM output in factual sources (USGS, FEMA, USDA) instead of raw model opinion, cutting hallucination risk in high-stakes farming decisions.
Mini-Interfaze: Receipt Field ExtractorLive | Code
PyTorch · CUDA · CRNN · Transformers · EasyOCR · Safetensors · Hugging Face Hub · Gradio
  • •Designed a native-fusion architecture from scratch: CRNN (CNN + BiLSTM + CTC) encoder → transformer decoder embedding space → fixed-schema JSON (company, date, address, total) with bounding-box + confidence metadata.
  • •Wrote a custom fused CUDA kernel (RMSNorm + residual) to speed up training.
Wispr Flow CloneLive | Code
PyTorch · CUDA · Transformers · Whisper · Hugging Face Hub · Gradio
  • •Fine-tuned a two-stage pipeline from scratch: Whisper ASR (verbatim, disfluency-preserving) → fine-tuned LLM for filler removal, grammar fixes, and tone adaptation across email/chat/notes modes, keeping all rewriting confined to an auditable system prompt.
  • •Wrote a custom fused CUDA kernel (bias-add + GELU) that reduces GPU memory traffic.
Message Notification RouterCode
Python · Sarvam AI (LLM/STT/Vision) · Pandas · Tenacity
  • •Built a multimodal WhatsApp router: text + transcribed voice (Saaras v3 STT) + OCR'd images (Sarvam Vision) → fused reasoning context → Sarvam-105B classifies notify/digest/mute.
  • •Designed a context-assembly layer fusing engagement, group roles, business history, and repetition signals into one JSON payload, feeding risk/scam signals as observations.

Technical Skills

Languages: Python, C++, CUDA, Bash, SQL

AI/ML: Machine Learning, Deep Learning, PyTorch, Hugging Face, Multimodal and Multilingual AI, Retrieval-Augmented Generation (RAG), Model Context Protocol (MCP), GPU Optimization, Fine-tuning (PEFT), Agentic AI, Prompt Engineering, Context Engineering

Frameworks: React, Flask, FastAPI, Express.js, REST APIs

Databases: MongoDB, Redis, Chroma DB

Tools: Git, GitHub, Docker, CI/CD, Jupyter Notebook

Achievements

  • •Ranked 1st in the Department of Artificial Intelligence & Data Science
  • •2nd Place at VNPS'25 – National-Level Project Showcase, VCET Mumbai
  • •Open-sourced models on Hugging Face Hub, collectively surpassing 100k+ downloads across Neural Networks, LLM's, ASR, translation, and vision architectures
  • •Launched aynlp, an open-source NLP utility library with 7k+ downloads, enabling developers to solve common NLP tasks efficiently
  • •Authored 3+ technical articles on AI Agents and Reinforcement Learning in VCET Techzette
  • •Kaggle competitor with 15+ badges across Notebooks and Competitions, applying data science and ML techniques to real-world problems
  • •Solved 350+ DSA problems across competitive coding platforms
  • •OSS Contribution basedosdados/sdk (feat(infra): (#1372) add GCP Console link), PR reviewed and merged by maintainers