How fast is the GetCred credit-scoring inference?

Sub-5ms zero-Python tree inference, validated on 2M+ simulated transactions, with zero raw financial statement retention under the DPDP Act 2023. The model is explainable, with SHAP attribution on each decision.

What latency problem does the Cache Cliff research address?

Edge AI model selection often optimizes for accuracy and overlooks the latency bottleneck on constrained hardware. The Cache Cliff framework benchmarks eight vision encoders on four edge platforms to make non-linear hardware trade-offs explicit.

How was real-time audio latency handled in Auda?

Auda used a native Opus codec pipeline with adaptive buffer pacing, designed around strict 20ms frame budgets and jitter-free 48kHz audio over WebSockets. It served 2.5 million cumulative listeners across 60,000+ communities.

How did the Medikabazaar payout pipeline improve?

Payout turnaround dropped from three days to under ten minutes across 400+ monthly transactions. The pipeline supported 10,000+ records across 28 states and improved downstream automation reliability by 35%.

Which systems skills show up across these projects?

Real-time streaming (WebSockets and Opus audio), low-latency model inference (sub-5ms tree inference and edge visual-language benchmarking), data pipelines for payouts and operations, and compliance-by-design under the DPDP Act 2023.

Selected Work | Yashvardhan Singh

Founder's Office · 2026

PineBrook digital platforms & workflow automation

Delivered web platforms (issafoundation.co.in, pinebrooktechnologies.com, specialsprings.com) and built an interactive demo & compliance automation system.

Shipped 3 production platforms, cut unqualified enterprise demos by 60%, shortened sales cycles by three weeks, and reduced compliance documentation effort by 80%.

Digital platformsProduct discoveryPrototypesWorkflow automation

Read case study Visit live project

Fintech Product · 2025

GetCred alternative credit intelligence

Explainable cashflow underwriting and alternative credit scoring for India’s 80M+ gig and informal workers.

Enabled instant 0–900 GigScore evaluation, sub-5ms decision latency on 2M+ simulated monthly transactions, and cryptographically auditable zero-retention reports.

Product discoveryXGBoost & SHAPUPI TelemetryDPDP Compliance

Read case study Visit live project

Operations systems · 2025

Medikabazaar operations data

Moved a payout workflow from multi-day manual handling to an operational system teams could trust.

Payout turnaround dropped from three days to under ten minutes across 400+ monthly transactions; 10,000+ records across 28 states supported a 35% reliability improvement downstream.

Data pipelinesOperations designReliability

Read case study Visit live project

Systems research · 2026

Cache Cliff edge-AI research

A hardware-aware way to choose vision encoders for edge visual-language systems.

Introduced the Cache Cliff terminology to make non-linear hardware trade-offs explicit; readers should validate recommendations on their target hardware.

BenchmarkingEdge inferenceLatency analysis

Read case study Visit live project

Audio Systems · 2021–2024

Auda (Latin: audiere / to hear) distributed VoIP audio pipeline & integrations

Real-time voice channel audio streaming engine serving 2.5 Million listeners over low-latency Opus codec with Discord Webhooks and API integrations.

Served 2.5 Million cumulative listeners and 60,000+ communities with jitter-free 48kHz Opus audio transmission, robust webhook dispatching, and high-throughput real-time voice channel integrations.

Node.jsOpus CodecDiscord API & WebhooksWebSocketsVoIP Pipelines

Read case study Visit live project

Frequently asked questions

How fast is the GetCred credit-scoring inference?: Sub-5ms zero-Python tree inference, validated on 2M+ simulated transactions, with zero raw financial statement retention under the DPDP Act 2023. The model is explainable, with SHAP attribution on each decision.
What latency problem does the Cache Cliff research address?: Edge AI model selection often optimizes for accuracy and overlooks the latency bottleneck on constrained hardware. The Cache Cliff framework benchmarks eight vision encoders on four edge platforms to make non-linear hardware trade-offs explicit.
How was real-time audio latency handled in Auda?: Auda used a native Opus codec pipeline with adaptive buffer pacing, designed around strict 20ms frame budgets and jitter-free 48kHz audio over WebSockets. It served 2.5 million cumulative listeners across 60,000+ communities.
How did the Medikabazaar payout pipeline improve?: Payout turnaround dropped from three days to under ten minutes across 400+ monthly transactions. The pipeline supported 10,000+ records across 28 states and improved downstream automation reliability by 35%.
Which systems skills show up across these projects?: Real-time streaming (WebSockets and Opus audio), low-latency model inference (sub-5ms tree inference and edge visual-language benchmarking), data pipelines for payouts and operations, and compliance-by-design under the DPDP Act 2023.

From problem and constraint to working system.

PineBrook digital platforms & workflow automation

GetCred alternative credit intelligence

Medikabazaar operations data

Cache Cliff edge-AI research

Auda (Latin: audiere / to hear) distributed VoIP audio pipeline & integrations

Frequently asked questions

From problem and constraint to working system.

PineBrook digital platforms & workflow automation

GetCred alternative credit intelligence

Medikabazaar operations data

Cache Cliff edge-AI research

Auda (Latin: audiere / to hear) distributed VoIP audio pipeline & integrations

Frequently asked questions