Naimish Singh — builds agentic AI & backend systems
AI & backend engineer
Third-year CS (AI & ML) student. I design and ship agentic-AI, RAG, and backend systems — voice agents, knowledge-graph retrieval, real-time services. Open to AI & backend roles.
Agentic AI, RAG, and backend.
From real-time voice agents to knowledge-graph retrieval and reliable services — AI that actually ships.
Agentic AI & voice
Real-time voice agents and LLM tool-calling over telephony — LiveKit, Deepgram, Groq, with barge-in and semantic turn detection.
Talk agents 02 BackendBackend & APIs
FastAPI services with auth, Postgres/Redis, idempotency, background workers, and tests that hold under load.
Talk backend 03 SystemsRAG & retrieval
Hybrid and graph RAG with Neo4j and Qdrant, rerankers, and citation-bound answers evaluated with RAGAS.
Talk RAG 04 SafetyLLM safety & eval
Prompt-injection screening, PII redaction, guardrails, and evaluation gates that keep AI outputs trustworthy.
Talk safetyAPIs,
data,
reliability.
Backend is where I spend most of my time: clean FastAPI services, solid data models, idempotent event handling, and systems that keep working after the demo.
- REST APIs and ServersRoutes, controllers, validation, errors, and predictable contracts.
- Authentication and AccessLogin flows, protected routes, sessions, roles, and safer defaults.
- Database DesignSchemas, relationships, migrations, queries, and practical data modeling.
- Integrations and JobsConnecting APIs, background tasks, file flows, and automation endpoints.
- Testing and DeploymentPytest coverage, Docker, and CI so core behavior ships with fewer surprises.
RAG,
agents,
voice.
My AI work centers on agentic systems and retrieval — LLM tool-calling, multi-agent routing, and RAG — served through clean, evaluated backend pipelines.
- Applied ML & TransformersFine-tuning and serving models (MuRIL, ViT, WavLM) with real evaluation and error analysis.
- LLM ApplicationsAssistants, retrieval flows, structured outputs, and useful prompt systems.
- AI Agents and ToolsSmall agents that call tools, handle state, and stay grounded in the task.
- Data and Reporting FlowsParsing inputs, extracting fields, summarizing, and moving records cleanly.
- Human-Reviewed AIKeeping review points where accuracy, context, or risk actually matters.
AI becomes useful when it is backed by solid systems.
- StudyML
- BuildAPIs
- Testoutputs
- Shipsystems
The goal is not only to understand models. It is to build the backend, data, and product layers that make AI features useful in real workflows.
Careva
A Hindi voice-first health helpline — finds nearby PHCs, explains govt schemes, prices generic medicines, and flags emergencies to call 108.
- Problem
- Health system hard to navigate
- Move
- Hindi voice agent over phone
- Result
- 🏆 2nd · VoiceForBharat
Truthy
A multimodal deepfake detector for voice, image, and video — fusing transformer models with biophysical signals like rPPG pulse.
- Problem
- Deepfakes hard to detect
- Move
- Neural models + forensic signals
- Result
- Explainable verdict + confidence
engineering, under the hood
Backend and applied-ML projects where the work lives in the architecture, not the screenshot. All open source — read the code.
AuRAG
A knowledge-graph RAG engine linking equipment, failures, and procedures in Neo4j, with a LangGraph supervisor routing questions to specialist agents that return cited answers.
URL Shortener API
A FastAPI URL shortener and developer API with Redis-cached redirects, an off-request-path analytics worker, dual JWT/API-key auth, and sliding-window rate limiting.
Sign Language Recognition
A real-time ASL fingerspelling recognizer and assistive communication app, using MediaPipe hand landmarks feeding a lightweight MLP classifier served through Flask.
Hinglish Sarcasm Detector
Detects sarcasm in Hindi-English code-mixed text with a fine-tuned MuRIL transformer and a calibrated decision threshold, served through a Streamlit app.
Copilot Claude Proxy
A local proxy that speaks the Anthropic API and forwards to GitHub Copilot's chat models — so Claude Code and other Anthropic clients run on your Copilot plan.
Linkloader
A privacy-focused media downloader — paste a social link and stream the direct file at max quality, with stealth routing and no accounts, tracking, or server-side storage.
hackathons & wins
Most of these systems were built and shipped under hackathon pressure — then kept going after the clock stopped.
-
1st
Swadhikaar — The Operating System for Indic Patient Care
🏆 1stHackMatrix 2.0·IIT Patna × Jilo Health·2026 FinalistTechExpo, Techniche·IIT Guwahati·2026 -
1st
Guardial — LLM defence layer
🏆 1stGenAI Hackathon·GDG DevFest Lucknow·2025 2ndInter-Institutional Idea Challenge·BIET Lucknow·2025 -
2nd
Careva — multilingual AI helpline
💰 2nd10 Days of AI Voice Agents·Murf.ai (VoiceForBharat)·2026
I build AI systems and the backend that makes them real.
I'm Naimish Singh, a third-year CS (AI & ML) student. I design and ship agentic-AI and backend systems — real-time voice agents, knowledge-graph RAG, LLM safety layers, and the APIs and data models that hold them together.
A lot of this started under hackathon deadlines and freelance builds through my own studio, BitCrafts. I'm open to AI and backend roles where I can build at this level full-time.
- Ship, then hardenGet the core path working, then make it correct under load, edge cases, and real data.
- Evaluated AIRAG and agents backed by citations, guardrails, and eval gates — not vibes.
- Honest proofEvery project here is live or open source. No fake seniority, no invented experience.
tools I build with
The stack behind the projects — the frameworks, data stores, and infrastructure I reach for across agentic AI and backend work.
from concept to working project
I turn ideas into working systems: understand the problem, scope it small, build the core path, break it on purpose, then keep what held.
-
Study
Break the topic down into fundamentals, examples, and the questions I need to answer by building.
-
Scope
Choose a project small enough to finish but real enough to expose design, data, and backend decisions.
-
Build
Implement the interface, API, data flow, model behavior, or automation until the core path works.
-
Debug
Test the unhappy paths, tighten the code, inspect outputs, and make the system easier to reason about.
-
Reflect
Document what worked, what broke, and which engineering skill the next project should strengthen.
Need someone who can actually ship AI systems?
I'm open to AI and backend roles, internships, and serious collaborations. If the work here fits what your team needs, let's talk.
Student today, but the projects are real and the code is open. That's the point of this portfolio.
where I am right now
No inflated title, no mystery. This is my current direction and the kind of opportunity I am working toward.
Do you have industry experience?
No full-time industry role yet — I'm a third-year student. What I do have is shipped projects, freelance AI builds through my own studio, and hackathon wins. That's the experience I bring.
What do you actually build?
Agentic-AI and backend systems: real-time voice agents, knowledge-graph RAG, LLM safety layers, payment and ledger services, and the APIs behind them — mostly Python, FastAPI, and the LangChain / LangGraph stack.
What roles are you aiming for?
AI Engineer, Applied AI / LLM, and Backend Developer roles — internship or full-time. Anywhere I can ship real systems and keep leveling up.
What should someone contact you about?
First-role opportunities, internships, AI or backend collaborations, or sharp feedback on the work. Serious conversations welcome.
What kind of projects do you like building?
Ones where the backend logic matters — voice agents, retrieval systems, agents that call tools, and services that stay correct under real load.
Where can I see your code?
GitHub is the best place: github.com/ace-ify — most projects here are open source. My résumé is linked at the top and bottom of this page.
How should I reach you?
Email me at 21naimish21@gmail.com or connect on LinkedIn at linkedin.com/in/naimish21.
Send a role, collaboration, or useful technical note.
I'm open to AI and backend roles, internships, and project collaborations — especially where I can ship real systems.