# Gitesh Pareek

Backend & AI Engineer | Delhi, India

building bits. shipping bytes.

APIs, AI agents, and the queues doing the heavy lifting.

I build the systems that keep working after the demo ends.

## Experience

### Planet Sustech
Software Development Engineer, Backend & AI | 2024-09 – present | Remote, India
- AI systems, end to end. Independently deployed multi-agent services on AWS ECS Fargate, with gateways for routing, authentication, and conversation state.
- Documents into structured data. Built Python and Celery pipelines using AWS Bedrock for bill and invoice classification, extraction, and human review, with queue-based workers outside the interactive API request path. Delivered retrieval and report comparisons across 1,000+ listed companies.
- Backends built for real workloads. Cut API response time by 50% through MongoDB aggregation and indexing. Moved I/O-bound Celery workers from prefork to gevent and used Redis Pub/Sub for real-time messaging.
- Multi-tenant products. Delivered tenant-aware access controls, Azure AD SSO, SSE streaming, and application email through AWS SES. Built a four-module assessment platform collecting data from 800+ partner organizations.

## Projects

### Think Voice
AI developer SDK & conversational CLI.
A framework-independent toolkit that connects AI to application actions through repository-owned TypeScript capabilities, workflows, and input/output contracts. A Gemini-powered CLI helps developers integrate it into their codebase.
Technologies: TypeScript, Node.js, Gemini API, JSON Schema, TypeScript Compiler API, pnpm
- Reviewed integrations. The CLI inspects developer-selected source files, proposes changes, and applies them through reviewed diffs, schema validation, and stale-file checks.
- Static workflow compilation. The TypeScript Compiler API parses declarative definitions without executing project code, validates dependency graphs, detects cycles, and checks workflow inputs against action schemas.
- A local developer workbench. Inspect workflows, action bindings, schemas, and source definitions with live file refresh and dry-run previews of exact execution inputs.
- Execution tested in a CRM sandbox. Confirmed multi-step actions exercise tenant-scoped permissions, optimistic concurrency, idempotent writes, partial-failure recovery, and reconciliation when responses are lost after successful commits.

### Prism
From a rough idea to a build-ready plan.
An AI workspace that refines requirements through conversation, extracts design tokens from references, and generates four UI mockups with engineering task breakdowns.
Technologies: Next.js, tRPC, PostgreSQL, OpenRouter
-  Task-specific model routing with automatic failover and prompt-cache optimization. Vision-based design extraction connects visual references to UI generation. Frontend and backend task plans export to Excel.
Also built with Upstash Redis.

### Document Intelligence
Ask better questions of complex documents.
A retrieval-grounded assistant that turns PDFs into searchable knowledge, with answers and comparisons across organizations and reporting years.
Technologies: TypeScript, LangChain, Qdrant, Valkey
-  A document pipeline handles PDF parsing, chunking, embeddings, and indexing. Retrieval supplies relevant context for grounded answers and cross-organization, year-over-year comparisons.
Built with Next.js, LangChain, Qdrant, and Valkey.

## Toolkit
- Languages: TypeScript, JavaScript, Python, SQL, Java, Go
- Web & backend: Next.js, Node.js, Express.js, FastAPI, tRPC, Celery, gevent
- AI & retrieval: LangChain, AWS Bedrock, OpenRouter, Gemini API, RAG, Multi-agent orchestration
- Developer tooling: TypeScript Compiler API, JSON Schema, pnpm
- Data & storage: PostgreSQL, MongoDB, Redis, Valkey, Qdrant, Upstash Redis
- Cloud & deployment: Docker, AWS ECS Fargate, EC2, AWS SES
- Systems & auth: Multi-tenant SaaS, Azure AD SSO, SSE streaming, Redis Pub/Sub, Tenant-aware access
- Architecture: Service decomposition, API gateways, Asynchronous processing, Conversation state

## Education
B.Tech. in Computer Science, Maharaja Agrasen Institute of Technology, Delhi. 2020-08 – 2024-08. CGPA: 9.3.

## Achievements
- 2nd / 320+ submissions — 48-hour MLH hackathon
- 500+ problems solved — LeetCode

## Contact
Think we could build something valuable together?
I bring backend and AI experience, curiosity, and a willingness to learn. You bring a challenge worth taking on. Let’s see where we can help each other.
Email: pareekgitesh89@gmail.com
[GitHub](https://github.com/GenMech)
[LinkedIn](https://www.linkedin.com/in/itsgitesh/)
[Résumé](https://drive.google.com/file/d/1LrBUD9cP0OwR_IFlbT6UCpHsKaK-jnqJ/view?usp=sharing)

Content updated: 2026-10-09.