I break things, experiment, rewrite, and push engineering boundaries.

Backend engineer who owns AI infrastructure end-to-end. I build systems that scale, systems that stay up, and systems where the engineering decisions actually matter. No handoffs, no tickets—if there's a problem, I fix it.

Currently at Remotus, building the entire AI backend alone: APIs, LLM pipelines, vector search, async infrastructure, you name it.

Multi-CSV ingestion → GraphQL schema → LLM-powered analytics

• Ingested unstructured data (CSV, PDF, sheets) and mapped to queryable GraphQL schema
• Built RAG pipeline achieving 98% query accuracy on structured data
• Auto-generated dashboards with custom insights and complex calculations
• Solved the hard part: normalizing messy real-world data so LLMs can reason over it

Tech: FastAPI, PostgreSQL, Qdrant, LangChain, GraphQL, AWS SageMaker

Because off-the-shelf solutions couldn't keep up

• Dropped response times from 5s → 2s through parallel streaming
• Built on async WebSocket architecture, handling high concurrency
• Real-time token streaming for better UX

Tech: FastAPI, WebSockets, async Python, event-driven architecture

Live trading signal generation using Fyers API

• Real-time options chain processing every 60 seconds
• Computed OI signals, PCR, IV skew, max pain, gamma blast detection
• Dashboard + React-based strike picker for options trading decisions
• Built because existing tools didn't give me the signal I needed

Tech: FastAPI, Fyers API, Redis, React, real-time data pipelines

Handling 10,000+ concurrent users on a single web-space

• Architected CRDT-based design using YJS for race condition elimination
• Solved data sync issues in high-concurrency environments
• Maintained 99.9% uptime under production load

Tech: WebSockets, CRDT, async Python, connection pooling

• Automated model training and hosting workflow
• 30% cost reduction through intelligent resource management
• End-to-end pipeline from data ingestion to deployment

Tech: AWS SageMaker, Python, MLflow, Unsloth

• Scaled Qdrant to 8M vectors maintaining 90%+ retrieval accuracy
• 50% reduction in storage footprint
• 30% faster semantic search through index tuning and chunking strategy

Tech: Qdrant, embedding optimization, hybrid search strategies

Languages: Python, JavaScript, Java, C++, Bash

Backend: FastAPI, Django, Node.js, async I/O, microservices

Databases: PostgreSQL, MongoDB, Neo4j, Qdrant, Redis, ClickHouse

AI/ML: LLMs (OpenAI, Anthropic, open-source), RAG, prompt engineering, function-calling, LangChain, vector databases, model serving

Infrastructure: AWS (EC2, SageMaker, S3), Docker, Kubernetes, CI/CD, Terraform

Distributed Systems: Event-driven architecture, async queues (BullMQ, Celery), WebSocket streaming, load balancing, connection pooling

Tools & Practices: Git, GitHub, system design, SOLID principles, API design (REST, GraphQL), observability, monitoring

• Agentic AI architectures — multi-step orchestration, tool use, state management
• LLMOps & evals — building evaluation frameworks for production LLM systems
• Advanced prompt engineering — reasoning, chain-of-thought, structured outputs
• Production AI systems — reliability, observability, cost optimization at scale

• Built a hierarchical agent workflow where parent tasks split into specialized sub-agents that further decompose work based on input
• Solved infinite context memory management in LLM pipelines (one of the hard problems nobody talks about)
• Optimized LLM API costs by 40% through intelligent caching, prompt compression, and request deduplication
• Designed event-driven streaming architectures with Kafka and Redis Streams for real-time inference

Actively contributing to projects that solve real problems:

• Cheshire Cat — Refactored backend endpoints, improved architecture
• db-agent — LLM integration and hallucination reduction (8+ PRs)
• db-pilot — LLM integration and model serving improvements
• GitHub Issue Metrics — Documentation refactoring
• PyVista — Fixed execution path issues
• SkriptLang — Updated build scripts and core infrastructure
• AcadVault — Backend improvements

Master's in Information Technology
Dhirubhai Ambani Institute of Information and Communication Technology (DAIICT)
CPI: 8.0/10.0 (Jul 2022 - Jun 2024)

Bachelor's in Computer Application
The Maharaja Sayajirao University of Baroda
CPI: 8.53/10.0 (Jul 2019 - May 2022)

• Systems, not features — Architecture decisions matter more than code velocity
• Ownership mindset — I debug at 2am because the problem exists, not because someone asked
• Production first — Code that doesn't run in production isn't code
• Measure before optimizing — I use metrics, logs, and profiling to find real bottlenecks
• Trade-offs over perfection — Fast and working beats slow and perfect
• Learn from others' code — Open source contributions teach faster than writing solo

✅ Systems I build stay up without constant babysitting
✅ Architecture decisions that compound in value over months
✅ Teammates saying "this code is easy to understand"
✅ Finding the bottleneck in 30 minutes instead of 3 hours
✅ Building something that didn't exist before

Backend Engineer | AI Infrastructure | LLM Integration roles at:

• Well-funded startups building AI systems (not just wrapping APIs)
• Product companies solving real problems at scale
• Organizations where engineering decisions matter

Ideal setup: Remote, fast-moving team, ownership from day one, hard problems that compound learning

Have an interesting problem? Building something ambitious? Hit me up.

• Email: mihirkohli2001@gmail.com
• LinkedIn: linkedin.com/in/mihirkohli
• GitHub: github.com/MihirKohli
• Phone: +91-9737635737