HELLO, I'M

Manish Sharma

Machine Learning Developer

Download Resume Contact Me
Manish Sharma

About Me

I make AI work for real and complex use cases. Not just flashy demos 🚀

Location: Bangalore, 560068

Work Experience

Machine Learning Engineer - 2

Parspec.io, Bangalore | Website | LinkedIn

Jan 2024 – Present

  • Engineered wireframes and architected EPICOR CRM Integration within Parspec Inventory PF Tools.
  • Reduced model number to datasheet mapping latency by 80% using Cache and Gemini 2.0 Flash LLM.
  • Fine-tuned Llama 3.3 70B Instruct Model on a custom dataset for attribute extraction (Modal Labs H100).
  • Developed an unstructured model-number to datasheet mapping system using heuristics and OSS LLMs.
  • Achieved 90% accuracy matching model numbers across 5M documents; 91% attribute extraction accuracy.
  • Enhanced header/column detection algorithm (4 to 8 columns, 97% accuracy), reducing manual processing by 30%.
  • Designed and deployed an order information detection algorithm (T5-base, 95% accuracy).

Machine Learning Scientist

Docsumo-AI, Bangalore | Website | LinkedIn

Dec 2022 – Dec 2023

  • Designed and implemented advanced Document KV + Table Extractor (LayoutLM, BROS, YOLO).
  • Integrated ML/DL architectures into 10+ Custom APIs for clients (MRR $80K-100K).
  • Built and integrated Chat-AI using LangChain and PineCone-DB for QA/support tasks.
  • Reduced annotation time from 1 day to ~2hrs using GPT-KV LLM Extractor (GPT-4).
  • Implemented advanced Synthetic Data Generation Pipeline (FP-Tree algos), outperforming production APIs.
  • Worked on KV/Checkbox Extraction for forms like Insurance (1040, 1120s, W9s), Bank Cheques.

Research Assistant Intern

Indian Institute of Science, Bangalore

June 2021 – Sep 2021

  • Collated and preprocessed massive Hindi datasets for OCR and Speech Recognition.
  • Leveraged PyTesseract and EasyOCR for text extraction; evaluated using WordLevel Accuracy.
  • Worked with Librosa and MelSpectrogram for ASR tasks.

Projects / Side Builds

V-Rag [Video Based RAG System]

Developed a system allowing users to query video content via YouTube URL or uploaded videos. Implemented video chunking/indexing with Qdrant, integrated a QA pipeline, experimented with vector databases, and deployed with Streamlit.

Github | Loom
QdrantRAGVideo-QueryingStreamlitVector DB

GeoPuzzle [Geography Quiz App]

Extracted data of 10,000+ global places using Selenium/BS4, cleaned/tagged data, added interactive game features, and built a UI using Lovable UI framework.

Game Link | Twitter
PILLovable UIWeb ScrapingSeleniumBeautifulSoup

AutoCommit Generator

Built an AutoCommit Generator using Ollama and Mistral for local commit message generation. Fully local, privacy-focused, simple bash script integration (<100 lines of code).

Github | Twitter
MistralOllamaGithubLLMBashLocal AI

Skills

Languages & Frameworks

Python Cursor FastAPI GitHub Ollama OpenRouter Lovable HuggingFace Grok

AI & Machine Learning
Machine Learning Deep Learning NLP PyTorch TensorFlow HuggingFace Transformers LLMs (GPT's) Finetuned Models RAG Systems Multi-Agent RAGs AI Agents

Natural Language Processing

Spacy SciSpacy MedXN Librosa PyTesseract

Databases & Analytics

MySQL Neo4j Tableau Amplitude Analytics Hasura-DB

Cloud & Collaboration

AWS GCP Google Colab Kubernetes Docker

Data Structures & Algorithms

Problem Solving Optimization Algorithm Design

Get In Touch

Interested in collaborating or have a question? Feel free to reach out!

Send Email Connect on LinkedIn

Location: Bangalore, 560068