Open to new opportunities

Senior Python Backend · AI / LLM Engineer

Rishi Prakash

I build production AI systems and high-performance backends in Python — real-time streaming, voice AI, and LLM integration that serve thousands of users at scale.

View my work Get in touch

6+: Years building
99.9%: Uptime shipped
<100ms: P95 latency

01 / About

From data science to AI engineering

I'm a Python backend engineer who came up through data science. Six years ago I was building predictive models and analytics dashboards for Fortune 500 clients — today I architect the backend for an enterprise AI coaching platform: real-time streaming conversations, voice AI, and retrieval-augmented knowledge systems used by thousands of people.

That data background still shapes how I build. I care about measurable outcomes, clean data flow, and systems that hold up under real load — whether that means keeping streaming responses under 100ms, integrating LLMs with function calling, or designing for 99.9% uptime.

I work primarily in Python — FastAPI, asyncio, Pydantic — alongside Azure OpenAI, Redis, and modern observability tooling. I'm equally happy owning a system end-to-end or going deep on a hard problem like prompt-injection defense or zero-downtime model migrations. Mostly, I like turning ambitious AI ideas into reliable software that ships.

02 / Skills

Skills & tooling

Languages & Frameworks

Python 3.11+
FastAPI
Pydantic v2
asyncio
httpx
SQL
C# / .NET (integration)

AI / ML & LLM

Azure OpenAI
OpenAI Realtime API
RAG & embeddings
Function calling
Prompt engineering
Injection defense

Real-time & Streaming

Server-Sent Events
WebRTC
LiveKit
Async generators
Voice AI

Data & Storage

Redis Vector Store
Redis caching
PostgreSQL
MySQL
pandas
NumPy

DevOps & Observability

Docker
Kubernetes
Azure
DataDog APM
Prometheus
pytest / pytest-asyncio

Foundations

System design
Predictive modeling
NLP / topic modeling
A/B testing
Data visualization

03 / Experience

Experience

2024 — Present

Senior Python Backend Engineer · Degreed

Lead backend engineer for an enterprise AI coaching platform, building the Python/FastAPI services behind real-time coaching, voice, and assessment experiences used by thousands of learners.
- Architected real-time AI conversations with FastAPI + Server-Sent Events, sustaining sub-100ms P95 streaming latency.
- Built a voice-coaching capability on WebRTC and the OpenAI Realtime API with robust connection and quality monitoring.
- Designed a retrieval-augmented knowledge base (vector search) so coaches answer from uploaded, domain-specific content.
- Hardened the platform with multi-layer prompt-injection defense and a structured exception/error-code system.
- Led a zero-downtime model migration with dual-model fallback, and cut Redis operations ~80% via unified session storage.
- FastAPI
- Azure OpenAI
- Redis
- LiveKit
- DataDog
Earlier

Data Scientist / Senior Data Analyst · LatentView Analytics

Delivered analytics and machine-learning solutions for Fortune 500 clients, turning messy data into decisions executives could act on.
- Built predictive models and ML pipelines for client business problems.
- Designed interactive Tableau and Qlik Sense dashboards for executive decision-making.
- Ran NLP topic modeling (LDA) over large feedback corpora to surface themes.
- Python
- SQL
- scikit-learn
- Tableau
- NLP
2018 — 2022

Data Science & Analytics Roles · Multiple organizations

Built the data foundation that still shapes how I engineer software today.
- Developed ETL pipelines and BI reporting for stakeholders.
- Performed statistical analysis and predictive modeling across domains.
- Python
- SQL
- Statistics
2018 — 2019

Business Analytics & Data Science · SSN School of Advanced Career Education (with IBM)

Residential program: business statistics, BI & visualization, optimization, big data (Hadoop, Spark), and predictive analytics. GPA 8.4/10.

04 / Work

Selected projects

Real-time AI Coaching Engine

Problem: Coaching conversations need to feel instant, personal, and stateful — even under heavy concurrent load.
Approach: Streaming responses over Server-Sent Events with async generators, per-user context injection, and Redis-backed session persistence.
Outcome: Sub-100ms P95 streaming latency for thousands of concurrent sessions.

FastAPI
Azure OpenAI
SSE
Redis

Voice Coaching Platform

Problem: Make coaching feel like a real spoken conversation, not a chatbot with text-to-speech bolted on.
Approach: Live voice sessions on WebRTC and the OpenAI Realtime API, with structured event handling, transcription capture, and quality monitoring.
Outcome: Natural, low-latency voice coaching with automatic reconnection and call-quality safeguards.

LiveKit
OpenAI Realtime
WebRTC

RAG Knowledge Base

Problem: Coaches must answer from an organization's own documents — accurately and without leaking sensitive data.
Approach: A document pipeline with semantic chunking, vector embeddings, and approximate-nearest-neighbor retrieval, plus PII detection and masking.
Outcome: Sub-second, domain-specific retrieval that meets enterprise compliance needs.

Vector Search
Embeddings
Redis
PII masking

Prompt-Injection Defense

Problem: Any user-generated content is an attack surface for prompt injection against the model.
Approach: LLM-powered injection detection layered with auth, group, and entity validation, backed by a structured exception and error-code system.
Outcome: Hardened every user-content path with defense-in-depth and clear, secure error handling.

Security
Azure OpenAI
Pydantic

Details are kept high-level to respect employer confidentiality — happy to walk through architecture and trade-offs in conversation.

05 / Impact

By the numbers

Years building software

0k+

Lines of production Python

API endpoints shipped

Uptime maintained

<0ms

P95 streaming latency

AI models integrated

06 / Contact

Let's build something

I'm open to backend and AI/LLM engineering roles where I can ship reliable systems that put AI to real use. If that sounds like your team, I'd love to talk.

Email me Download résumé