Poojith Mapakshi profile photo 1 on Tal

Poojith Mapakshi

Software Engineer | LLM & Backend Systems | Distributed Processing | FastAPI Celery | Python | Building scalable ML pipelines at Straive

Puducherry

6+ years of work experience

Straive Technologies
Apr 20242 yrs 5 mosPresent

Architected scalable ML platform supporting real-time LLM inference and production agentic workflows — owned full product lifecycle from scoping requirements to deployment and continuous iteration. Designed and deployed multi-agent AI system processing 800+ documents daily; integrated Gemini 2.5 and GPT-4 with production safety guarantees and quality evaluation pipelines. Championed model quality and integrity — built multi-layer QA infrastructure (pattern guards, entity validation, semantic verification) achieving 95% reduction in faulty LLM outputs in production. MLOps & SDLC best practices: Implemented CI/CD pipelines (GitHub Actions, Docker, Kubernetes); maintained model versioning, automated testing, and zero-downtime deployments in production. Drove innovation in ML scalability: Redesigned async LLM orchestration with AsyncIO and parallel Celery workers — 65% latency reduction and 60% throughput increase at zero infrastructure cost. Production reliability: Architected fault-tolerant system with idempotent consumers, retry mechanisms, safe fallbacks, and real-time monitoring achieving 99% uptime under operational load. Data infrastructure: Optimized PostgreSQL for high-concurrency AI workloads (connection pooling, query optimization, 40% DB load reduction); built Redis-backed caching and event-driven pipelines for big data processing.
Straive
Feb 20242 yrs 7 mosPresent

• Architected Doc Prep—an internal prompt-engineering automation system—reducing manual structuring by 80% and scaling to 300+ manuscripts/day • Built multi-layer GuardRails (numerical, proper nouns, scientific/medical terms, UK/US spelling), and integrated QualityAgentCheck using GPT-4o/Gemini; reduced faulty edits by 95% • Engineered high-scale backend + LLM pipelines using FastAPI, Celery, Redis, async workers, and autoscaling strategies • Solved a production bottleneck caused by 80–90 redundant LLM calls per table; redesigned orchestration to cut processing time from 5–6 mins → <2 mins (65% faster) • Implemented text-comparison evaluation pipelines for entity, numerical, and factual consistency—improving editorial quality by 25% • Tuned Celery concurrency + Redis caching to reduce latency by 40% and push reliability toward 99.99% uptime • Built real-time observability dashboards for latency, worker activity, queue depth, and failures
Straive
Feb 20243 mosApr 2024

Puducherry, India

• Improved marketing decision-making by developing a segmentation model that increased conversions by 15% • Designed the initial prototype of Doc Prep, enabling 80% reduction in manual document structuring • Enhanced ML reliability by optimizing CI/CD for model versioning, cutting deployment time from hours to minutes
Puducherry Technological University
Jun 20204 yrs 3 mosAug 2024

United Kingdom

Poojith’s AI Stack

aistudio
aistudio
antigravity
antigravity
codex
codex
cursor
cursor
hugging face
hugging face
Figma Make
Figma Make

Education

Puducherry Technological University

Bachelor of Engineering, Computer Engineering

Puducherry Technological University

Join Poojith and 10k+ other professionals on

Tal Boss

Tal is the app where bosses meet engineers & builders directly, without middlemen

Download Tal Boss

Poojith Mapakshi on Tal

Headline

Software Engineer | LLM & Backend Systems | Distributed Processing | FastAPI Celery | Python | Building scalable ML pipelines at Straive

About

contact: poojithmapakshi2@gmail.com AI/ML Engineer and Backend Developer My work is mainly on building scalable LLM systems, distributed processing workflows, and high-reliability backend services. I use Python, FastAPI, Celery, Redis, Spark, Databricks and common NLP and LLM tools. I am open to roles in AI/ML Engineering, LLM systems, Backend Engineering and applied ML. My work includes developing FastAPI-based microservices, high-concurrency Celery pipelines, Redis-backed caching layers, and evaluation frameworks for numerical, entity, and factual consistency. Technical experience spans FastAPI, Python, Celery, Redis, Spark, Databricks, PyTorch, HuggingFace, NLP workflows, distributed systems, and large-scale text processing. Open to roles in AI/ML Engineering, LLM Systems, Backend Engineering, NLP, and Applied ML.

Location

Puducherry

AI stack

aistudio

antigravity

codex

cursor

hugging face

Figma Make

Work experience

Senior Software Engineer at Straive Technologies, Apr 2024 - Present · 2 yrs 5 mos, Architected scalable ML platform supporting real-time LLM inference and production agentic workflows — owned full product lifecycle from scoping requirements to deployment and continuous iteration. Designed and deployed multi-agent AI system processing 800+ documents daily; integrated Gemini 2.5 and GPT-4 with production safety guarantees and quality evaluation pipelines. Championed model quality and integrity — built multi-layer QA infrastructure (pattern guards, entity validation, semantic verification) achieving 95% reduction in faulty LLM outputs in production. MLOps & SDLC best practices: Implemented CI/CD pipelines (GitHub Actions, Docker, Kubernetes); maintained model versioning, automated testing, and zero-downtime deployments in production. Drove innovation in ML scalability: Redesigned async LLM orchestration with AsyncIO and parallel Celery workers — 65% latency reduction and 60% throughput increase at zero infrastructure cost. Production reliability: Architected fault-tolerant system with idempotent consumers, retry mechanisms, safe fallbacks, and real-time monitoring achieving 99% uptime under operational load. Data infrastructure: Optimized PostgreSQL for high-concurrency AI workloads (connection pooling, query optimization, 40% DB load reduction); built Redis-backed caching and event-driven pipelines for big data processing.

Software Engineer at Straive, Feb 2024 - Present · 2 yrs 7 mos, • Architected Doc Prep—an internal prompt-engineering automation system—reducing manual structuring by 80% and scaling to 300+ manuscripts/day • Built multi-layer GuardRails (numerical, proper nouns, scientific/medical terms, UK/US spelling), and integrated QualityAgentCheck using GPT-4o/Gemini; reduced faulty edits by 95% • Engineered high-scale backend + LLM pipelines using FastAPI, Celery, Redis, async workers, and autoscaling strategies • Solved a production bottleneck caused by 80–90 redundant LLM calls per table; redesigned orchestration to cut processing time from 5–6 mins → <2 mins (65% faster) • Implemented text-comparison evaluation pipelines for entity, numerical, and factual consistency—improving editorial quality by 25% • Tuned Celery concurrency + Redis caching to reduce latency by 40% and push reliability toward 99.99% uptime • Built real-time observability dashboards for latency, worker activity, queue depth, and failures

Software Engineer Intern at Straive, Feb 2024 - Apr 2024 · 3 mos, Puducherry, India, • Improved marketing decision-making by developing a segmentation model that increased conversions by 15% • Designed the initial prototype of Doc Prep, enabling 80% reduction in manual document structuring • Enhanced ML reliability by optimizing CI/CD for model versioning, cutting deployment time from hours to minutes

Education at Puducherry Technological University, Jun 2020 - Aug 2024 · 4 yrs 3 mos, United Kingdom

Education

Bachelor of Engineering, Computer Engineering, Puducherry Technological University