pankaj = {
    "role":        "AI + Backend Engineer",
    "building":    "Manthan AI — 13-project production AI ecosystem",
    "stack":       ["LangGraph", "FastAPI", "Node.js", "Docker", "PostgreSQL", "Redis"],
    "expertise":   ["RAG Systems", "Agentic AI", "LLM Fine-Tuning", "AI Evaluation"],
    "location":    "Delhi NCR, India",
    "open_to":     "AI / GenAI Internships 2027 · Remote · OSS Collaboration",
    "philosophy":  "If it can't be deployed, monitored, evaluated, and scaled — it isn't finished.",
}

Most AI engineers understand models. The rare ones can build the complete system around the model — retrieval, orchestration, evaluation, infrastructure, and deployment. That's where I operate.

AI-powered research and knowledge intelligence platform with enterprise-grade retrieval architecture.

Architecture: Hybrid BM25 + Dense Retrieval → Cross-Encoder Reranking → LLM Generation → Ragas Eval Pipeline

Stack:

Key decisions:

• Hybrid BM25 + dense retrieval over naive vector search — keyword recall matters for domain-specific corpora
• Ragas evaluation pipeline integrated into CI/CD — faithfulness gates every deployment
• FastAPI with SSE streaming, JWT auth, rate limiting — not Streamlit

Autonomous research system with stateful multi-step reasoning, tool orchestration, and memory-aware workflows.

Agent Loop: Plan → Tool Call → Error Recovery → State Update → Memory Write → Next Step

Capabilities:

• Multi-step reasoning with persistent state graphs (LangGraph)
• Tool registry: web search, code execution, retrieval, structured output
• Short-term context + long-term vector memory
• Streaming FastAPI backend with real-time agent thought visibility
• Full Docker + GitHub Actions CI/CD pipeline

Stack:

End-to-end fine-tuning pipeline: data curation → LoRA training → quantization → production serving.

Stack:

Production ML monitoring system with automated drift detection, retraining triggers, and experiment tracking.

Data → FastAPI Endpoint → MLflow Tracking → Evidently AI Drift → GitHub Actions Retrain → Model Registry

Stack:

 

A unified platform of 13 interconnected AI projects — not a portfolio of demos, but a production-grade AI engineering ecosystem. Each project exposes a FastAPI layer, is containerized with Docker, and feeds signal into the next.

manthan-ai/
├── scholarforge-ai/          ← Production RAG + Evaluation  [SHIPPED]
├── multi-agent-research/     ← LangGraph Agentic System     [SHIPPED]
├── domain-llm-pipeline/      ← LoRA Fine-Tune + Serving     [SHIPPED]
├── mlops-drift-pipeline/     ← MLflow + Evidently AI        [SHIPPED]
├── ai-saas-platform/         ← Real users, full-stack AI    [IN PROGRESS]
├── llm-eval-framework/       ← Custom DeepEval harness      [PLANNED]
├── code-review-agent/        ← LangGraph + GitHub API       [PLANNED]
└── ...6 more                 ← Evolution Engine             [ROADMAP]

The evolution engine: Each project feeds a critic-agent feedback loop — prompts evolve, architectures refine, and evaluation scores compound across the ecosystem.

EdTech startup · Delhi · Full-time

• Joined as one of the earliest engineering members; shaped both product and backend architecture from the ground up
• Designed and built REST API infrastructure with Node.js + Express; integrated MongoDB for content management at scale
• Built and shipped end-to-end features: user authentication, content delivery pipelines, internal CMS
• Developed the initial React Native mobile application MVP
• Contributed to system design decisions targeting 10K+ student users
• Worked directly with founders — product thinking, cross-functional execution, zero-to-one delivery

Node.js Express MongoDB React Native REST APIs System Design

   

Open to: AI/GenAI internship opportunities · OSS collaboration · Referrals · Technical conversations

"If it can't be deployed, monitored, evaluated, and scaled — it isn't finished."