This video can't be played inline here.
Watch it directly ↗
A global professional services and consulting firm's Innovation and Business team was manually evaluating "value pools," business and product ideas submitted by employees, researching each one by hand across 12 evaluation dimensions before producing a readiness score. They partnered with Lyzr to build a parallel-sequential agent workflow that automates the process end to end.
Each submitted value pool required in-depth manual research across 12 dimensions, from market sizing to strategic alignment, before a single overall readiness score and per-dimension scores could be produced.
Without a standardized process, the manual approach introduced variability in evaluation quality from one reviewer to the next, and the effort involved slowed down decision-making on which ideas to pursue.
The time required to research and score each value pool by hand limited how many ideas the business team could realistically evaluate in depth.
Lyzr built the Vera Research Agent, a parallel-sequential, orchestrated agent workflow that automates value pool intake, research, scoring, and reporting end to end.
Employees submit value pools through a guided Quick Form or a pre-filled upload template, capturing structured input across 17 fields, with real-time validation, and select which of the 12 evaluation dimensions they want assessed.
Each evaluation dimension is handled by a pair of specialized agents, a research agent that performs web-based research and a validation agent that structures the findings into JSON, with all selected dimension pairs running in parallel to maximize throughput.
Each dimension is scored out of 4 using predefined weighted logic, normalized for consistency, and rolled up into a single consolidated readiness score for every value pool.
Once all selected dimensions are researched and scored, a set of report agents compiles an executive summary, dimension-wise insights, market opportunity assessment, risk indicators, and strategic recommendations, stored and retrievable via API.
Every value pool now moves through the same defined dimensions and the same weighted scoring model, replacing evaluation quality that previously varied by reviewer.
Automating research and scoring removes the manual bottleneck that previously capped how many value pools could be evaluated in depth at once.
Parallel agent research across dimensions is designed to move a value pool from submission to a scored, reportable output far faster than manual desk research.
Generated reports are retrievable by value pool ID via API, letting employees and reviewers pull structured evaluation data without waiting on manual compilation.
The platform runs on AWS within a dedicated VPC, with a WAF and load balancer fronting public traffic, EKS-hosted frontend and backend services in private app subnets, and IAM, Secrets Manager, and KMS governing access and credentials.
Value pool submissions, dimension scores, and generated reports are stored in Amazon DocumentDB within private DB subnets, isolated from direct internet access.
AWS CloudTrail logs all API calls and account activity, giving the business team a compliant, traceable record of how every value pool was submitted, researched, and scored.
Every deployment in this case study started with one conversation. Bring your use case, your stack, your constraints. We'll show you the path from here.
