Pranjal is wanted for orchestrating autonomous AI systems and full-stack architectures end to end. Authorities report the suspect was last seen deploying high-impact neural pipelines while actively remaining on the run for remote AI/ML engineering roles.
Field intelligence reveals a habitual pattern of converting complex research papers into production-ready web APIs, autonomous multi-agent workflows, and scalable backends. Investigators urge engineering teams to approach with competitive remote opportunities before another offer claims the subject.
Bystanders describe the suspect as armed with strong coffee, dangerously clicky keyboard, and an unyielding refusal to leave a bug unresolved overnight. Approach with caution.
He likes the whole pipeline from a LangGraph node diagram on paper to a webhook firing in production.
Evaluating plate images of 4 waterborne pathogens E. coli, K. pneumoniae, P. aeruginosa and S. aureus with CNN models (ResNet and EfficientNet) and Vision Transformer models (Swin T and ConvNeXt). Finding their model perfoemances and evaluating robustness and generalization capabilities across different datasets and conditions. This will be converted into a usable product after the paper is published.
AI-powered web app that tailors resumes and cover letters to job descriptions in seconds, built with FastAPI, LangGraph, PostgreSQL and Next.js. Focus is solely on the anti fabrication mechanism implemented. Designed an 8-node LLM pipeline with ID-based fact matching to rewrite only relevant resume content, integrated Google OAuth, AWS S3 storage, and Razorpay billing for a production-ready SaaS product.
Built a ReAct agent loop in LangGraph with conditional tool-calling edges across web search, filesystem, GitHub, and Gmail via MCP. Persisted multi-turn conversation history via async SQLite checkpointer; Streamlit frontend with live streaming and tool-status indicators.
Built a GPT-2 style decoder-only transformer in PyTorch (10.8M and 25M params) with self-attention, residual connections, and layer norm from scratch. Curated 52M characters from 13 pathology textbooks, cleaned OCR artifacts, and published dataset on HuggingFace (with 59 public downloads). Achieved val loss 0.8794 on the 25M model.
Built a stacked ensemble (Random Forest, XGBoost, LightGBM, Logistic Regression) on highly imbalanced data using class weighting and undersampling. Achieved F1 of 0.85 (Precision: 0.92, Recall: 0.79) after threshold tuning; deployed as a FastAPI endpoint on Render.
| Category | Code | Substances (Tools & Tech) | Finding |
|---|---|---|---|
| Languages | LANG | Python · SQL · JavaScript | Primary tool |
| Agentic AI Frameworks | AG-AI | LangChain · LangGraph · LangSmith · MCP | Core Focus |
| ML / Deep Learning | ML-DL | PyTorch · scikit-learn · XGBoost · LightGBM | Core Focus |
| Web Development | WEB | FastAPI · Django · Next.js · Streamlit · Tailwind | Primary tool |
| Data Tools | DATA | Pandas · NumPy · Matplotlib · Seaborn | Comfortable |
| Databases | DB | PostgreSQL · MongoDB · SQLite | Comfortable |
| Deployment | DEPL | Railway · Render · Vercel · AWS S3 | Comfortable |
| Developer Tools | DEV | Git · Docker · Linux · Google Colab · Kaggle · Google OAuth | Comfortable |
An internship, a role, or a good question about agentic systems send it through and he'll get back to you.