Hi, I’m
GANESHA A M
AI developer
Bangalore,KA
Summary
Forward Deployed AI Engineer building production-grade multi-agent systems, RAG architectures, and autonomous AI applications. Spearheaded notYC (notycombinator.com)-an AI-powered YC application reviewer that received direct feedback from Garry Tan, CEO of Y Combinator. Expertise in LangGraph state machines, vector database optimization, cost-efficient inference routing, and end-to-end product ownership across data engineering, LLM orchestration, and deployment infrastructure.
Overview
1
year of professional experience
1
Certification
Work History
notYCBangalore
AI Product Engineer (Founding Engineer Intern)
01.2026 - 05.2026
Job overview
- Architected end-to-end RAG pipeline ingesting 100GB+ YC data via recursive chunking (8-10K tokens) into ChromaDB with batch reranking, reducing review latency from 3 minutes to 45 seconds.
- Engineered retrieval-based feedback engine using LangChain + Claude 3.5 Sonnet for section-by-section analysis, achieving 87% founder satisfaction across 500+ applications with zero manual review.
- Designed data pipeline (Python + FastAPI) streaming YC data through LLM entity extraction, validation, and PostgreSQL persistence, processing 10K+ documents weekly at 99.2% accuracy.
- Optimized inference costs via async batch processing and smart routing, reducing per-request LLM calls by 40%, cutting operational cost from $0.45 to $0.27 per application.
- Early-stage AI Startup
- Bangalore
CapgeminiBangalore
AI/LLM Intern
08.2025 - 11.2025
Job overview
- Completed 3-month curriculum covering LLM fundamentals, agentic AI patterns (ReAct, tool use, agent memory), prompt engineering, and core data structures, earning Capgemini certification.
- Capgemini
- Bangalore
Education
Christ Academy Institute of Advanced Studies
Master of Computer Applications
01.2026
University Overview
Davangere Institute of Advanced Management Studies
Bachelor of Computer Applications
08.2023
University Overview
Skills
- Multi-Agent Orchestration
- Agent-to-Agent Protocol (A2A)
- ReAct
- Tool Use / Function Calling
- LangGraph State Machines
- Agent Memory
- Guardrails & Safety
- Planning & Task Decomposition
- Reflection & Self-Correction
- Agent Evaluation & Benchmarking
- LangSmith Observability
- Retrieval-Augmented Generation (RAG)
- Hybrid Search (Dense Sparse)
- Vector Database Optimization
- Fine-tuning (QLoRA, LoRA, Unsloth)
- Prompt Engineering
- Few-Shot Learning
- Chain-of-Thought
- Multi-Modal Processing
- System 1 / System 2 Routing
- Synthetic Data Generation
- Model Context Protocol (MCP)
- LangChain
- LangGraph
- LangSmith
- LlamaIndex
- Pinecone
- ChromaDB
- DeepAgents SDK
- Browser-use
- FastMCP
- Google ADK
- Anthropic SDK
- Claude (35 Sonnet, 3 Opus)
- Gemini (20 Flash, 20 Flash Thinking, 15 Flash)
- Llama 32
- GPT-4 Turbo
- Hugging Face Transformers
- FastAPI
- PostgreSQL (pgvector)
- Redis
- Docker
- Render
- Hugging Face Spaces
- Railway
- Playwright
- Scikit-learn
- Python
- SQL
- JavaScript / TypeScript
- YAML
Certification
- Advanced MCP (Anthropic)
- Sub-Agents (Anthropic)
- Machine Learning (Infosys)
- LLM & Agentic AI (Capgemini)
Hackathons And Recognition
Hackathons And Recognition
- Meta AI India Hackathon, Phase 2 Finalist, Built custom RL simulator (OpenEnv + Python) evaluating autonomous agent decisions across 3 financial task tiers. Scored 9.5/10 in Phase 1, advancing as Phase 2 Finalist.
- Aptos Hackathon, Recognition from Judges, Shipped AI-guided crypto wallet app (React, Node.js) in 48 hours across 10+ blockchains. Received recognition certificate from Aptos judges.
Projects
Projects
- Security Auditing SLM, Live, Fine-tuned Llama 3.2 1B using Unsloth + QLoRA, achieving 80%+ factual grounding accuracy and reducing model size by 75% for on-device deployment., Engineered secure FastAPI backend with Argon2 hashing, Redis rate limiting, handling 50+ concurrent users (99.8% uptime) integrated with ChromaDB and PostgreSQL., Implemented self-correction loop using LangChain + FastMCP, removing 92% false positives. Generated 64-char API keys for third-party integrations (parity with Gemini/Anthropic APIs).
- Sovereign Intelligence Agent, Live, Architected System 1 / System 2 routing using LiteLLM, routing 65% to Gemini 2.0 Flash-Lite (4x lower latency), reducing end-to-end latency from 8s to 3.2s while maintaining 94% answer quality., Integrated live web browsing via DeepAgents SDK + Playwright, handling 1000+ monthly queries at 96.4% retrieval accuracy across multimodal content (text + images + tables)., Built validation framework with synthetic data factory, Judge LLM grading, LangSmith tracing, and 3-point pre-flight checks reducing deployment crashes from 3.2% to 0%.
- Ledger AI - Multi-Agent Financial Platform, Live, Orchestrated LangGraph state machine (4 nodes: Planner, Researcher, Analyst, Auditor), reducing query-to-analysis time from 6 minutes to 2.1 minutes via parallel execution and state caching., Engineered Hybrid Search RAG combining BM25 + LlamaIndex + Pinecone, achieving 91% retrieval precision on financial queries, outperforming single-method baselines by 22%., Designed Semantic Caching system preventing 47% duplicate queries from hitting LLM, reducing costs by 31% and latency by 2.8 seconds. Implemented NER guardrails for PII masking.
Timeline
AI Product Engineer (Founding Engineer Intern)
notYC
01.2026 - 05.2026
AI/LLM Intern
Capgemini
08.2025 - 11.2025
Christ Academy Institute of Advanced Studies
Master of Computer Applications
Davangere Institute of Advanced Management Studies
Bachelor of Computer Applications