Engineering case studies: solving scale, delivery and technology problems

The challenge

The platform needed a distributed backend capable of supporting high-capacity video sessions and real-time state synchronization. The engineering challenge was to keep interactive media traffic from competing with core application transactions while maintaining reliable synchronization with external calendars.

What Hithika did

The architecture separated media routing from transactional services, introduced service discovery and gateway-based routing, and used Kafka for event-driven synchronization. The Jitsi media layer was horizontally scaled so video workloads could grow independently from core application services.

Business outcome

• 100,000 concurrent live video sessions supported
• 5 core platform modules delivered from the ground up
• 60% reduction in enterprise calendar meeting collisions
• Sub-second transmission performance for live sessions

The challenge

A constrained local talent market was slowing product delivery and creating a growing backlog. The new team also had to fit the client's existing engineering practices rather than introduce a parallel delivery model.

What Hithika did

The team reworked the central Researcher Portal around clearer domain boundaries, automated unit testing and the client's existing CI/CD patterns. Delivery was managed through metric-based Agile sprints so the new engineers could become productive without a long transition period.

Business outcome

• Specialized engineering group operational within 10 business days
• 40% increase in sprint velocity
• Six-month product backlog cleared ahead of schedule
• Full-capacity engineering output reached within the first week

The challenge

Monolithic query patterns and persistent HTTP polling created unnecessary load and contributed to system-wide latency. Peak traffic exposed the limits of the existing architecture and put user experience at risk.

What Hithika did

A distributed experimentation and routing layer was introduced for high-throughput traffic. Persistent polling was replaced with a multi-node WebSocket pipeline, while Redis absorbed high-frequency reads and reduced repeated pressure on the primary data stores.

Business outcome

• Sustained 250–300 TPS under persistent traffic
• 70% reduction in profile processing and transmission latency
• Staged feature releases without exposing the entire user base to structural changes

The challenge

Users had to complete too many manual steps before the platform could build a useful profile. At the same time, document ingestion services had to cope with irregular files and unstructured resume content.

What Hithika did

Resume parsing was integrated behind an elastic Spring Cloud Gateway and moved into asynchronous processing. MongoDB was used for flexible document data while MySQL retained structured, searchable profile information. A resume builder reduced the amount of manual data entry required.

Business outcome

• 90% increase in resume/document creation
• 70% reduction in registration processing time
• 60% improvement in profile completeness

The challenge

Manual validation meant processing capacity was directly tied to the size of the operations team. Peak campaigns created queues and made scaling expensive.

What Hithika did

The ingestion workflow was redesigned around AWS Lambda and API Gateway so compute could scale with demand. Automated extraction and validation logic was introduced to classify incoming documents and extract required data in real time.

Business outcome

• 100% automated ingestion and validation workflow
• Reduced dependence on manual review for routine processing
• Elastic compute model aligned infrastructure cost with actual processing demand
• Capacity could grow without a linear increase in operations headcount

The challenge

Keyword-based recommendations could match literal terms but struggled to understand relationships between skills, experience and learning content. The platform needed contextual matching without sacrificing response time.

What Hithika did

Machine-learning features were designed to represent relationships between user skills and content. TensorFlow-based models were combined with an ultra-low-latency Aerospike feature store so recommendation vectors could be retrieved and evaluated quickly.

Business outcome

• Real-time contextual skill and course recommendations
• <15 ms profile-matching processing window
• More relevant career and development recommendations to support content adoption

The challenge

Memory leaks, background thread issues and failures during asynchronous network operations were causing production crashes. The engineering team also needed stronger release guardrails to prevent regressions.

What Hithika did

The mobile runtime was audited to isolate memory and threading problems. Automated quality controls were added to the GitLab CI/CD pipeline, and performance monitoring was used to track stability through subsequent releases.

Business outcome

• Crash sessions reduced from 0.35% to 0.05%
• 85% reduction in crash rate
• Improved release predictability and production confidence
• Lower risk of user churn caused by application instability

The challenge

Large serial jobs were pushing processing into the following business day. Database contention and monolithic SQL routines limited the ability to increase throughput safely.

What Hithika did

Processing was partitioned into parallel worker pipelines using Spring Batch and multi-node execution. ActiveMQ introduced asynchronous processing between stages, reducing serial dependencies and spreading workload across workers.

Business outcome

• 30% reduction in end-of-day processing time
• Reduced risk of batch processing spilling into the next business day
• Lower database contention and better use of existing infrastructure

The challenge

The notification path could not absorb peak bursts reliably. When traffic increased, messages were delayed or dropped, weakening the connection between listing activity and customer action.

What Hithika did

Kafka was introduced as a durable event buffer so traffic spikes could be absorbed without dropping notifications. A distributed MongoDB layer provided fast access to notification template data while the messaging services scaled independently.

Business outcome

• 35% increase in system-driven conversion
• Zero message drops reported across high-volume notification bursts
• Notification infrastructure became more resilient to peak traffic
• Optimized customer touchpoints contributed materially to revenue performance

The challenge

Engineering priorities were competing, technical debt was slowing delivery and security concerns were beginning to affect enterprise sales conversations. The organization needed experienced technical direction without immediately committing to a full-time CTO structure.

What Hithika did

A technology and infrastructure assessment identified architectural bottlenecks, technical debt and security gaps. The work combined architecture planning, VAPT review, engineering process changes and metric-driven delivery management. A practical multi-year technology roadmap was then created around the business's growth requirements.

Business outcome

• Delivered a practical 3-year technology roadmap
• Delivery milestone accuracy improved from approximately 60% to 95%
• Critical infrastructure vulnerabilities identified and addressed
• Stronger security readiness for enterprise customer conversations